Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mount.h> |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
07f3f05c | 40 | #include "internal.h" |
b2dba1af | 41 | #include "mount.h" |
1da177e4 | 42 | |
789680d1 NP |
43 | /* |
44 | * Usage: | |
873feea0 NP |
45 | * dcache->d_inode->i_lock protects: |
46 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
47 | * dcache_hash_bucket lock protects: |
48 | * - the dcache hash table | |
49 | * s_anon bl list spinlock protects: | |
50 | * - the s_anon list (see __d_drop) | |
19156840 | 51 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
52 | * - the dcache lru lists and counters |
53 | * d_lock protects: | |
54 | * - d_flags | |
55 | * - d_name | |
56 | * - d_lru | |
b7ab39f6 | 57 | * - d_count |
da502956 | 58 | * - d_unhashed() |
2fd6b7f5 NP |
59 | * - d_parent and d_subdirs |
60 | * - childrens' d_child and d_parent | |
b23fb0a6 | 61 | * - d_alias, d_inode |
789680d1 NP |
62 | * |
63 | * Ordering: | |
873feea0 | 64 | * dentry->d_inode->i_lock |
b5c84bf6 | 65 | * dentry->d_lock |
19156840 | 66 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
67 | * dcache_hash_bucket lock |
68 | * s_anon lock | |
789680d1 | 69 | * |
da502956 NP |
70 | * If there is an ancestor relationship: |
71 | * dentry->d_parent->...->d_parent->d_lock | |
72 | * ... | |
73 | * dentry->d_parent->d_lock | |
74 | * dentry->d_lock | |
75 | * | |
76 | * If no ancestor relationship: | |
789680d1 NP |
77 | * if (dentry1 < dentry2) |
78 | * dentry1->d_lock | |
79 | * dentry2->d_lock | |
80 | */ | |
fa3536cc | 81 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
82 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
83 | ||
74c3cbe3 | 84 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 85 | |
949854d0 | 86 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 87 | |
e18b890b | 88 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 89 | |
232d2d60 WL |
90 | /** |
91 | * read_seqbegin_or_lock - begin a sequence number check or locking block | |
92 | * lock: sequence lock | |
93 | * seq : sequence number to be checked | |
94 | * | |
95 | * First try it once optimistically without taking the lock. If that fails, | |
96 | * take the lock. The sequence number is also used as a marker for deciding | |
97 | * whether to be a reader (even) or writer (odd). | |
98 | * N.B. seq must be initialized to an even number to begin with. | |
99 | */ | |
100 | static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq) | |
101 | { | |
48f5ec21 | 102 | if (!(*seq & 1)) /* Even */ |
232d2d60 | 103 | *seq = read_seqbegin(lock); |
48f5ec21 | 104 | else /* Odd */ |
232d2d60 WL |
105 | write_seqlock(lock); |
106 | } | |
107 | ||
48f5ec21 | 108 | static inline int need_seqretry(seqlock_t *lock, int seq) |
232d2d60 | 109 | { |
48f5ec21 AV |
110 | return !(seq & 1) && read_seqretry(lock, seq); |
111 | } | |
112 | ||
113 | static inline void done_seqretry(seqlock_t *lock, int seq) | |
114 | { | |
115 | if (seq & 1) | |
232d2d60 | 116 | write_sequnlock(lock); |
232d2d60 WL |
117 | } |
118 | ||
1da177e4 LT |
119 | /* |
120 | * This is the single most critical data structure when it comes | |
121 | * to the dcache: the hashtable for lookups. Somebody should try | |
122 | * to make this good - I've just made it work. | |
123 | * | |
124 | * This hash-function tries to avoid losing too many bits of hash | |
125 | * information, yet avoid using a prime hash-size or similar. | |
126 | */ | |
127 | #define D_HASHBITS d_hash_shift | |
128 | #define D_HASHMASK d_hash_mask | |
129 | ||
fa3536cc ED |
130 | static unsigned int d_hash_mask __read_mostly; |
131 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 132 | |
b07ad996 | 133 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 134 | |
8966be90 | 135 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 136 | unsigned int hash) |
ceb5bdc2 | 137 | { |
6d7d1a0d LT |
138 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
139 | hash = hash + (hash >> D_HASHBITS); | |
ceb5bdc2 NP |
140 | return dentry_hashtable + (hash & D_HASHMASK); |
141 | } | |
142 | ||
1da177e4 LT |
143 | /* Statistics gathering. */ |
144 | struct dentry_stat_t dentry_stat = { | |
145 | .age_limit = 45, | |
146 | }; | |
147 | ||
3942c07c | 148 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 149 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
150 | |
151 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
152 | |
153 | /* | |
154 | * Here we resort to our own counters instead of using generic per-cpu counters | |
155 | * for consistency with what the vfs inode code does. We are expected to harvest | |
156 | * better code and performance by having our own specialized counters. | |
157 | * | |
158 | * Please note that the loop is done over all possible CPUs, not over all online | |
159 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
160 | * on and off. If one of them goes off, we will just keep their counters. | |
161 | * | |
162 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
163 | * please update all vfs counters to match. | |
164 | */ | |
3942c07c | 165 | static long get_nr_dentry(void) |
3e880fb5 NP |
166 | { |
167 | int i; | |
3942c07c | 168 | long sum = 0; |
3e880fb5 NP |
169 | for_each_possible_cpu(i) |
170 | sum += per_cpu(nr_dentry, i); | |
171 | return sum < 0 ? 0 : sum; | |
172 | } | |
173 | ||
62d36c77 DC |
174 | static long get_nr_dentry_unused(void) |
175 | { | |
176 | int i; | |
177 | long sum = 0; | |
178 | for_each_possible_cpu(i) | |
179 | sum += per_cpu(nr_dentry_unused, i); | |
180 | return sum < 0 ? 0 : sum; | |
181 | } | |
182 | ||
312d3ca8 CH |
183 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
184 | size_t *lenp, loff_t *ppos) | |
185 | { | |
3e880fb5 | 186 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 187 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 188 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
189 | } |
190 | #endif | |
191 | ||
5483f18e LT |
192 | /* |
193 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
194 | * The strings are both count bytes long, and count is non-zero. | |
195 | */ | |
e419b4cc LT |
196 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
197 | ||
198 | #include <asm/word-at-a-time.h> | |
199 | /* | |
200 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
201 | * aligned allocation for this particular component. We don't | |
202 | * strictly need the load_unaligned_zeropad() safety, but it | |
203 | * doesn't hurt either. | |
204 | * | |
205 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
206 | * need the careful unaligned handling. | |
207 | */ | |
94753db5 | 208 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 209 | { |
bfcfaa77 | 210 | unsigned long a,b,mask; |
bfcfaa77 LT |
211 | |
212 | for (;;) { | |
12f8ad4b | 213 | a = *(unsigned long *)cs; |
e419b4cc | 214 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
215 | if (tcount < sizeof(unsigned long)) |
216 | break; | |
217 | if (unlikely(a != b)) | |
218 | return 1; | |
219 | cs += sizeof(unsigned long); | |
220 | ct += sizeof(unsigned long); | |
221 | tcount -= sizeof(unsigned long); | |
222 | if (!tcount) | |
223 | return 0; | |
224 | } | |
225 | mask = ~(~0ul << tcount*8); | |
226 | return unlikely(!!((a ^ b) & mask)); | |
e419b4cc LT |
227 | } |
228 | ||
bfcfaa77 | 229 | #else |
e419b4cc | 230 | |
94753db5 | 231 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 232 | { |
5483f18e LT |
233 | do { |
234 | if (*cs != *ct) | |
235 | return 1; | |
236 | cs++; | |
237 | ct++; | |
238 | tcount--; | |
239 | } while (tcount); | |
240 | return 0; | |
241 | } | |
242 | ||
e419b4cc LT |
243 | #endif |
244 | ||
94753db5 LT |
245 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
246 | { | |
6326c71f | 247 | const unsigned char *cs; |
94753db5 LT |
248 | /* |
249 | * Be careful about RCU walk racing with rename: | |
250 | * use ACCESS_ONCE to fetch the name pointer. | |
251 | * | |
252 | * NOTE! Even if a rename will mean that the length | |
253 | * was not loaded atomically, we don't care. The | |
254 | * RCU walk will check the sequence count eventually, | |
255 | * and catch it. And we won't overrun the buffer, | |
256 | * because we're reading the name pointer atomically, | |
257 | * and a dentry name is guaranteed to be properly | |
258 | * terminated with a NUL byte. | |
259 | * | |
260 | * End result: even if 'len' is wrong, we'll exit | |
261 | * early because the data cannot match (there can | |
262 | * be no NUL in the ct/tcount data) | |
263 | */ | |
6326c71f LT |
264 | cs = ACCESS_ONCE(dentry->d_name.name); |
265 | smp_read_barrier_depends(); | |
266 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
267 | } |
268 | ||
9c82ab9c | 269 | static void __d_free(struct rcu_head *head) |
1da177e4 | 270 | { |
9c82ab9c CH |
271 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
272 | ||
b3d9b7a3 | 273 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
1da177e4 LT |
274 | if (dname_external(dentry)) |
275 | kfree(dentry->d_name.name); | |
276 | kmem_cache_free(dentry_cache, dentry); | |
277 | } | |
278 | ||
279 | /* | |
b5c84bf6 | 280 | * no locks, please. |
1da177e4 LT |
281 | */ |
282 | static void d_free(struct dentry *dentry) | |
283 | { | |
0d98439e | 284 | BUG_ON((int)dentry->d_lockref.count > 0); |
3e880fb5 | 285 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
286 | if (dentry->d_op && dentry->d_op->d_release) |
287 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 288 | |
dea3667b LT |
289 | /* if dentry was never visible to RCU, immediate free is OK */ |
290 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 291 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 292 | else |
9c82ab9c | 293 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
294 | } |
295 | ||
31e6b01f NP |
296 | /** |
297 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 298 | * @dentry: the target dentry |
31e6b01f NP |
299 | * After this call, in-progress rcu-walk path lookup will fail. This |
300 | * should be called after unhashing, and after changing d_inode (if | |
301 | * the dentry has not already been unhashed). | |
302 | */ | |
303 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
304 | { | |
305 | assert_spin_locked(&dentry->d_lock); | |
306 | /* Go through a barrier */ | |
307 | write_seqcount_barrier(&dentry->d_seq); | |
308 | } | |
309 | ||
1da177e4 LT |
310 | /* |
311 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
312 | * d_iput() operation if defined. Dentry has no refcount |
313 | * and is unhashed. | |
1da177e4 | 314 | */ |
858119e1 | 315 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 316 | __releases(dentry->d_lock) |
873feea0 | 317 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
318 | { |
319 | struct inode *inode = dentry->d_inode; | |
320 | if (inode) { | |
321 | dentry->d_inode = NULL; | |
b3d9b7a3 | 322 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 323 | spin_unlock(&dentry->d_lock); |
873feea0 | 324 | spin_unlock(&inode->i_lock); |
f805fbda LT |
325 | if (!inode->i_nlink) |
326 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
327 | if (dentry->d_op && dentry->d_op->d_iput) |
328 | dentry->d_op->d_iput(dentry, inode); | |
329 | else | |
330 | iput(inode); | |
331 | } else { | |
332 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
333 | } |
334 | } | |
335 | ||
31e6b01f NP |
336 | /* |
337 | * Release the dentry's inode, using the filesystem | |
338 | * d_iput() operation if defined. dentry remains in-use. | |
339 | */ | |
340 | static void dentry_unlink_inode(struct dentry * dentry) | |
341 | __releases(dentry->d_lock) | |
873feea0 | 342 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
343 | { |
344 | struct inode *inode = dentry->d_inode; | |
345 | dentry->d_inode = NULL; | |
b3d9b7a3 | 346 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
347 | dentry_rcuwalk_barrier(dentry); |
348 | spin_unlock(&dentry->d_lock); | |
873feea0 | 349 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
350 | if (!inode->i_nlink) |
351 | fsnotify_inoderemove(inode); | |
352 | if (dentry->d_op && dentry->d_op->d_iput) | |
353 | dentry->d_op->d_iput(dentry, inode); | |
354 | else | |
355 | iput(inode); | |
356 | } | |
357 | ||
da3bbdd4 | 358 | /* |
dd1f6b2e | 359 | * dentry_lru_(add|del|move_list) must be called with d_lock held. |
da3bbdd4 KM |
360 | */ |
361 | static void dentry_lru_add(struct dentry *dentry) | |
362 | { | |
8aab6a27 | 363 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) { |
19156840 | 364 | spin_lock(&dentry->d_sb->s_dentry_lru_lock); |
8aab6a27 | 365 | dentry->d_flags |= DCACHE_LRU_LIST; |
a4633357 CH |
366 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
367 | dentry->d_sb->s_nr_dentry_unused++; | |
62d36c77 | 368 | this_cpu_inc(nr_dentry_unused); |
19156840 | 369 | spin_unlock(&dentry->d_sb->s_dentry_lru_lock); |
a4633357 | 370 | } |
da3bbdd4 KM |
371 | } |
372 | ||
23044507 NP |
373 | static void __dentry_lru_del(struct dentry *dentry) |
374 | { | |
375 | list_del_init(&dentry->d_lru); | |
dd1f6b2e | 376 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
23044507 | 377 | dentry->d_sb->s_nr_dentry_unused--; |
62d36c77 | 378 | this_cpu_dec(nr_dentry_unused); |
23044507 NP |
379 | } |
380 | ||
f0023bc6 SW |
381 | /* |
382 | * Remove a dentry with references from the LRU. | |
dd1f6b2e DC |
383 | * |
384 | * If we are on the shrink list, then we can get to try_prune_one_dentry() and | |
385 | * lose our last reference through the parent walk. In this case, we need to | |
386 | * remove ourselves from the shrink list, not the LRU. | |
f0023bc6 | 387 | */ |
da3bbdd4 KM |
388 | static void dentry_lru_del(struct dentry *dentry) |
389 | { | |
dd1f6b2e DC |
390 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
391 | list_del_init(&dentry->d_lru); | |
392 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; | |
393 | return; | |
394 | } | |
395 | ||
da3bbdd4 | 396 | if (!list_empty(&dentry->d_lru)) { |
19156840 | 397 | spin_lock(&dentry->d_sb->s_dentry_lru_lock); |
23044507 | 398 | __dentry_lru_del(dentry); |
19156840 | 399 | spin_unlock(&dentry->d_sb->s_dentry_lru_lock); |
da3bbdd4 KM |
400 | } |
401 | } | |
402 | ||
b48f03b3 | 403 | static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list) |
da3bbdd4 | 404 | { |
dd1f6b2e DC |
405 | BUG_ON(dentry->d_flags & DCACHE_SHRINK_LIST); |
406 | ||
19156840 | 407 | spin_lock(&dentry->d_sb->s_dentry_lru_lock); |
a4633357 | 408 | if (list_empty(&dentry->d_lru)) { |
8aab6a27 | 409 | dentry->d_flags |= DCACHE_LRU_LIST; |
b48f03b3 | 410 | list_add_tail(&dentry->d_lru, list); |
a4633357 | 411 | } else { |
b48f03b3 | 412 | list_move_tail(&dentry->d_lru, list); |
dd1f6b2e DC |
413 | dentry->d_sb->s_nr_dentry_unused--; |
414 | this_cpu_dec(nr_dentry_unused); | |
da3bbdd4 | 415 | } |
19156840 | 416 | spin_unlock(&dentry->d_sb->s_dentry_lru_lock); |
da3bbdd4 KM |
417 | } |
418 | ||
d52b9086 MS |
419 | /** |
420 | * d_kill - kill dentry and return parent | |
421 | * @dentry: dentry to kill | |
ff5fdb61 | 422 | * @parent: parent dentry |
d52b9086 | 423 | * |
31f3e0b3 | 424 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
425 | * |
426 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 427 | * |
b5c84bf6 NP |
428 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
429 | * d_kill. | |
d52b9086 | 430 | */ |
2fd6b7f5 | 431 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 432 | __releases(dentry->d_lock) |
2fd6b7f5 | 433 | __releases(parent->d_lock) |
873feea0 | 434 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 435 | { |
d52b9086 | 436 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
437 | /* |
438 | * Inform try_to_ascend() that we are no longer attached to the | |
439 | * dentry tree | |
440 | */ | |
b161dfa6 | 441 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
442 | if (parent) |
443 | spin_unlock(&parent->d_lock); | |
d52b9086 | 444 | dentry_iput(dentry); |
b7ab39f6 NP |
445 | /* |
446 | * dentry_iput drops the locks, at which point nobody (except | |
447 | * transient RCU lookups) can reach this dentry. | |
448 | */ | |
d52b9086 | 449 | d_free(dentry); |
871c0067 | 450 | return parent; |
d52b9086 MS |
451 | } |
452 | ||
c6627c60 DH |
453 | /* |
454 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
455 | * dentry->d_lock is locked. The caller must take care of that, if | |
456 | * appropriate. | |
457 | */ | |
458 | static void __d_shrink(struct dentry *dentry) | |
459 | { | |
460 | if (!d_unhashed(dentry)) { | |
461 | struct hlist_bl_head *b; | |
462 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
463 | b = &dentry->d_sb->s_anon; | |
464 | else | |
465 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
466 | ||
467 | hlist_bl_lock(b); | |
468 | __hlist_bl_del(&dentry->d_hash); | |
469 | dentry->d_hash.pprev = NULL; | |
470 | hlist_bl_unlock(b); | |
471 | } | |
472 | } | |
473 | ||
789680d1 NP |
474 | /** |
475 | * d_drop - drop a dentry | |
476 | * @dentry: dentry to drop | |
477 | * | |
478 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
479 | * be found through a VFS lookup any more. Note that this is different from | |
480 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
481 | * possible, giving a successful _negative_ lookup, while d_drop will | |
482 | * just make the cache lookup fail. | |
483 | * | |
484 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
485 | * reason (NFS timeouts or autofs deletes). | |
486 | * | |
487 | * __d_drop requires dentry->d_lock. | |
488 | */ | |
489 | void __d_drop(struct dentry *dentry) | |
490 | { | |
dea3667b | 491 | if (!d_unhashed(dentry)) { |
c6627c60 | 492 | __d_shrink(dentry); |
dea3667b | 493 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
494 | } |
495 | } | |
496 | EXPORT_SYMBOL(__d_drop); | |
497 | ||
498 | void d_drop(struct dentry *dentry) | |
499 | { | |
789680d1 NP |
500 | spin_lock(&dentry->d_lock); |
501 | __d_drop(dentry); | |
502 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
503 | } |
504 | EXPORT_SYMBOL(d_drop); | |
505 | ||
77812a1e NP |
506 | /* |
507 | * Finish off a dentry we've decided to kill. | |
508 | * dentry->d_lock must be held, returns with it unlocked. | |
509 | * If ref is non-zero, then decrement the refcount too. | |
510 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
511 | */ | |
dd1f6b2e DC |
512 | static inline struct dentry * |
513 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | |
77812a1e NP |
514 | __releases(dentry->d_lock) |
515 | { | |
873feea0 | 516 | struct inode *inode; |
77812a1e NP |
517 | struct dentry *parent; |
518 | ||
873feea0 NP |
519 | inode = dentry->d_inode; |
520 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e | 521 | relock: |
dd1f6b2e DC |
522 | if (unlock_on_failure) { |
523 | spin_unlock(&dentry->d_lock); | |
524 | cpu_relax(); | |
525 | } | |
77812a1e NP |
526 | return dentry; /* try again with same dentry */ |
527 | } | |
528 | if (IS_ROOT(dentry)) | |
529 | parent = NULL; | |
530 | else | |
531 | parent = dentry->d_parent; | |
532 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
533 | if (inode) |
534 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
535 | goto relock; |
536 | } | |
31e6b01f | 537 | |
0d98439e LT |
538 | /* |
539 | * The dentry is now unrecoverably dead to the world. | |
540 | */ | |
541 | lockref_mark_dead(&dentry->d_lockref); | |
542 | ||
f0023bc6 | 543 | /* |
f0023bc6 SW |
544 | * inform the fs via d_prune that this dentry is about to be |
545 | * unhashed and destroyed. | |
546 | */ | |
590fb51f | 547 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
548 | dentry->d_op->d_prune(dentry); |
549 | ||
550 | dentry_lru_del(dentry); | |
77812a1e NP |
551 | /* if it was on the hash then remove it */ |
552 | __d_drop(dentry); | |
553 | return d_kill(dentry, parent); | |
554 | } | |
555 | ||
1da177e4 LT |
556 | /* |
557 | * This is dput | |
558 | * | |
559 | * This is complicated by the fact that we do not want to put | |
560 | * dentries that are no longer on any hash chain on the unused | |
561 | * list: we'd much rather just get rid of them immediately. | |
562 | * | |
563 | * However, that implies that we have to traverse the dentry | |
564 | * tree upwards to the parents which might _also_ now be | |
565 | * scheduled for deletion (it may have been only waiting for | |
566 | * its last child to go away). | |
567 | * | |
568 | * This tail recursion is done by hand as we don't want to depend | |
569 | * on the compiler to always get this right (gcc generally doesn't). | |
570 | * Real recursion would eat up our stack space. | |
571 | */ | |
572 | ||
573 | /* | |
574 | * dput - release a dentry | |
575 | * @dentry: dentry to release | |
576 | * | |
577 | * Release a dentry. This will drop the usage count and if appropriate | |
578 | * call the dentry unlink method as well as removing it from the queues and | |
579 | * releasing its resources. If the parent dentries were scheduled for release | |
580 | * they too may now get deleted. | |
1da177e4 | 581 | */ |
1da177e4 LT |
582 | void dput(struct dentry *dentry) |
583 | { | |
8aab6a27 | 584 | if (unlikely(!dentry)) |
1da177e4 LT |
585 | return; |
586 | ||
587 | repeat: | |
98474236 | 588 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 589 | return; |
1da177e4 | 590 | |
8aab6a27 LT |
591 | /* Unreachable? Get rid of it */ |
592 | if (unlikely(d_unhashed(dentry))) | |
593 | goto kill_it; | |
594 | ||
595 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 596 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 597 | goto kill_it; |
1da177e4 | 598 | } |
265ac902 | 599 | |
39e3c955 | 600 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 601 | dentry_lru_add(dentry); |
265ac902 | 602 | |
98474236 | 603 | dentry->d_lockref.count--; |
61f3dee4 | 604 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
605 | return; |
606 | ||
d52b9086 | 607 | kill_it: |
dd1f6b2e | 608 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
609 | if (dentry) |
610 | goto repeat; | |
1da177e4 | 611 | } |
ec4f8605 | 612 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
613 | |
614 | /** | |
615 | * d_invalidate - invalidate a dentry | |
616 | * @dentry: dentry to invalidate | |
617 | * | |
618 | * Try to invalidate the dentry if it turns out to be | |
619 | * possible. If there are other dentries that can be | |
620 | * reached through this one we can't delete it and we | |
621 | * return -EBUSY. On success we return 0. | |
622 | * | |
623 | * no dcache lock. | |
624 | */ | |
625 | ||
626 | int d_invalidate(struct dentry * dentry) | |
627 | { | |
628 | /* | |
629 | * If it's already been dropped, return OK. | |
630 | */ | |
da502956 | 631 | spin_lock(&dentry->d_lock); |
1da177e4 | 632 | if (d_unhashed(dentry)) { |
da502956 | 633 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
634 | return 0; |
635 | } | |
636 | /* | |
637 | * Check whether to do a partial shrink_dcache | |
638 | * to get rid of unused child entries. | |
639 | */ | |
640 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 641 | spin_unlock(&dentry->d_lock); |
1da177e4 | 642 | shrink_dcache_parent(dentry); |
da502956 | 643 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
644 | } |
645 | ||
646 | /* | |
647 | * Somebody else still using it? | |
648 | * | |
649 | * If it's a directory, we can't drop it | |
650 | * for fear of somebody re-populating it | |
651 | * with children (even though dropping it | |
652 | * would make it unreachable from the root, | |
653 | * we might still populate it if it was a | |
654 | * working directory or similar). | |
50e69630 AV |
655 | * We also need to leave mountpoints alone, |
656 | * directory or not. | |
1da177e4 | 657 | */ |
98474236 | 658 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 659 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 660 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
661 | return -EBUSY; |
662 | } | |
663 | } | |
664 | ||
665 | __d_drop(dentry); | |
666 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
667 | return 0; |
668 | } | |
ec4f8605 | 669 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 670 | |
b5c84bf6 | 671 | /* This must be called with d_lock held */ |
dc0474be | 672 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 673 | { |
98474236 | 674 | dentry->d_lockref.count++; |
23044507 NP |
675 | } |
676 | ||
dc0474be | 677 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 678 | { |
98474236 | 679 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
680 | } |
681 | ||
b7ab39f6 NP |
682 | struct dentry *dget_parent(struct dentry *dentry) |
683 | { | |
df3d0bbc | 684 | int gotref; |
b7ab39f6 NP |
685 | struct dentry *ret; |
686 | ||
df3d0bbc WL |
687 | /* |
688 | * Do optimistic parent lookup without any | |
689 | * locking. | |
690 | */ | |
691 | rcu_read_lock(); | |
692 | ret = ACCESS_ONCE(dentry->d_parent); | |
693 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
694 | rcu_read_unlock(); | |
695 | if (likely(gotref)) { | |
696 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
697 | return ret; | |
698 | dput(ret); | |
699 | } | |
700 | ||
b7ab39f6 | 701 | repeat: |
a734eb45 NP |
702 | /* |
703 | * Don't need rcu_dereference because we re-check it was correct under | |
704 | * the lock. | |
705 | */ | |
706 | rcu_read_lock(); | |
b7ab39f6 | 707 | ret = dentry->d_parent; |
a734eb45 NP |
708 | spin_lock(&ret->d_lock); |
709 | if (unlikely(ret != dentry->d_parent)) { | |
710 | spin_unlock(&ret->d_lock); | |
711 | rcu_read_unlock(); | |
b7ab39f6 NP |
712 | goto repeat; |
713 | } | |
a734eb45 | 714 | rcu_read_unlock(); |
98474236 WL |
715 | BUG_ON(!ret->d_lockref.count); |
716 | ret->d_lockref.count++; | |
b7ab39f6 | 717 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
718 | return ret; |
719 | } | |
720 | EXPORT_SYMBOL(dget_parent); | |
721 | ||
1da177e4 LT |
722 | /** |
723 | * d_find_alias - grab a hashed alias of inode | |
724 | * @inode: inode in question | |
32ba9c3f LT |
725 | * @want_discon: flag, used by d_splice_alias, to request |
726 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
727 | * |
728 | * If inode has a hashed alias, or is a directory and has any alias, | |
729 | * acquire the reference to alias and return it. Otherwise return NULL. | |
730 | * Notice that if inode is a directory there can be only one alias and | |
731 | * it can be unhashed only if it has no children, or if it is the root | |
732 | * of a filesystem. | |
733 | * | |
21c0d8fd | 734 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
735 | * any other hashed alias over that one unless @want_discon is set, |
736 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 737 | */ |
32ba9c3f | 738 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 739 | { |
da502956 | 740 | struct dentry *alias, *discon_alias; |
1da177e4 | 741 | |
da502956 NP |
742 | again: |
743 | discon_alias = NULL; | |
b67bfe0d | 744 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 745 | spin_lock(&alias->d_lock); |
1da177e4 | 746 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 747 | if (IS_ROOT(alias) && |
da502956 | 748 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 749 | discon_alias = alias; |
32ba9c3f | 750 | } else if (!want_discon) { |
dc0474be | 751 | __dget_dlock(alias); |
da502956 NP |
752 | spin_unlock(&alias->d_lock); |
753 | return alias; | |
754 | } | |
755 | } | |
756 | spin_unlock(&alias->d_lock); | |
757 | } | |
758 | if (discon_alias) { | |
759 | alias = discon_alias; | |
760 | spin_lock(&alias->d_lock); | |
761 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
762 | if (IS_ROOT(alias) && | |
763 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 764 | __dget_dlock(alias); |
da502956 | 765 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
766 | return alias; |
767 | } | |
768 | } | |
da502956 NP |
769 | spin_unlock(&alias->d_lock); |
770 | goto again; | |
1da177e4 | 771 | } |
da502956 | 772 | return NULL; |
1da177e4 LT |
773 | } |
774 | ||
da502956 | 775 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 776 | { |
214fda1f DH |
777 | struct dentry *de = NULL; |
778 | ||
b3d9b7a3 | 779 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 780 | spin_lock(&inode->i_lock); |
32ba9c3f | 781 | de = __d_find_alias(inode, 0); |
873feea0 | 782 | spin_unlock(&inode->i_lock); |
214fda1f | 783 | } |
1da177e4 LT |
784 | return de; |
785 | } | |
ec4f8605 | 786 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
787 | |
788 | /* | |
789 | * Try to kill dentries associated with this inode. | |
790 | * WARNING: you must own a reference to inode. | |
791 | */ | |
792 | void d_prune_aliases(struct inode *inode) | |
793 | { | |
0cdca3f9 | 794 | struct dentry *dentry; |
1da177e4 | 795 | restart: |
873feea0 | 796 | spin_lock(&inode->i_lock); |
b67bfe0d | 797 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 798 | spin_lock(&dentry->d_lock); |
98474236 | 799 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
800 | /* |
801 | * inform the fs via d_prune that this dentry | |
802 | * is about to be unhashed and destroyed. | |
803 | */ | |
804 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
805 | !d_unhashed(dentry)) | |
806 | dentry->d_op->d_prune(dentry); | |
807 | ||
dc0474be | 808 | __dget_dlock(dentry); |
1da177e4 LT |
809 | __d_drop(dentry); |
810 | spin_unlock(&dentry->d_lock); | |
873feea0 | 811 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
812 | dput(dentry); |
813 | goto restart; | |
814 | } | |
815 | spin_unlock(&dentry->d_lock); | |
816 | } | |
873feea0 | 817 | spin_unlock(&inode->i_lock); |
1da177e4 | 818 | } |
ec4f8605 | 819 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
820 | |
821 | /* | |
77812a1e NP |
822 | * Try to throw away a dentry - free the inode, dput the parent. |
823 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
824 | * Releases dentry->d_lock. | |
d702ccb3 | 825 | * |
77812a1e | 826 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 827 | */ |
dd1f6b2e | 828 | static struct dentry * try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 829 | __releases(dentry->d_lock) |
1da177e4 | 830 | { |
77812a1e | 831 | struct dentry *parent; |
d52b9086 | 832 | |
dd1f6b2e | 833 | parent = dentry_kill(dentry, 0); |
d52b9086 | 834 | /* |
77812a1e NP |
835 | * If dentry_kill returns NULL, we have nothing more to do. |
836 | * if it returns the same dentry, trylocks failed. In either | |
837 | * case, just loop again. | |
838 | * | |
839 | * Otherwise, we need to prune ancestors too. This is necessary | |
840 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
841 | * is also expected to be beneficial in reducing dentry cache | |
842 | * fragmentation. | |
d52b9086 | 843 | */ |
77812a1e | 844 | if (!parent) |
dd1f6b2e | 845 | return NULL; |
77812a1e | 846 | if (parent == dentry) |
dd1f6b2e | 847 | return dentry; |
77812a1e NP |
848 | |
849 | /* Prune ancestors. */ | |
850 | dentry = parent; | |
d52b9086 | 851 | while (dentry) { |
98474236 | 852 | if (lockref_put_or_lock(&dentry->d_lockref)) |
dd1f6b2e DC |
853 | return NULL; |
854 | dentry = dentry_kill(dentry, 1); | |
d52b9086 | 855 | } |
dd1f6b2e | 856 | return NULL; |
1da177e4 LT |
857 | } |
858 | ||
3049cfe2 | 859 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 860 | { |
da3bbdd4 | 861 | struct dentry *dentry; |
da3bbdd4 | 862 | |
ec33679d NP |
863 | rcu_read_lock(); |
864 | for (;;) { | |
ec33679d NP |
865 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
866 | if (&dentry->d_lru == list) | |
867 | break; /* empty */ | |
868 | spin_lock(&dentry->d_lock); | |
869 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
870 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
871 | continue; |
872 | } | |
873 | ||
dd1f6b2e DC |
874 | /* |
875 | * The dispose list is isolated and dentries are not accounted | |
876 | * to the LRU here, so we can simply remove it from the list | |
877 | * here regardless of whether it is referenced or not. | |
878 | */ | |
879 | list_del_init(&dentry->d_lru); | |
880 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; | |
881 | ||
1da177e4 LT |
882 | /* |
883 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 884 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 885 | */ |
98474236 | 886 | if (dentry->d_lockref.count) { |
da3bbdd4 | 887 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
888 | continue; |
889 | } | |
ec33679d | 890 | rcu_read_unlock(); |
77812a1e | 891 | |
dd1f6b2e | 892 | dentry = try_prune_one_dentry(dentry); |
77812a1e | 893 | |
ec33679d | 894 | rcu_read_lock(); |
dd1f6b2e DC |
895 | if (dentry) { |
896 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
897 | list_add(&dentry->d_lru, list); | |
898 | spin_unlock(&dentry->d_lock); | |
899 | } | |
da3bbdd4 | 900 | } |
ec33679d | 901 | rcu_read_unlock(); |
3049cfe2 CH |
902 | } |
903 | ||
904 | /** | |
b48f03b3 DC |
905 | * prune_dcache_sb - shrink the dcache |
906 | * @sb: superblock | |
907 | * @count: number of entries to try to free | |
908 | * | |
909 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
910 | * done when we need more memory an called from the superblock shrinker | |
911 | * function. | |
3049cfe2 | 912 | * |
b48f03b3 DC |
913 | * This function may fail to free any resources if all the dentries are in |
914 | * use. | |
3049cfe2 | 915 | */ |
0a234c6d | 916 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan) |
3049cfe2 | 917 | { |
3049cfe2 CH |
918 | struct dentry *dentry; |
919 | LIST_HEAD(referenced); | |
920 | LIST_HEAD(tmp); | |
0a234c6d | 921 | long freed = 0; |
3049cfe2 | 922 | |
23044507 | 923 | relock: |
19156840 | 924 | spin_lock(&sb->s_dentry_lru_lock); |
3049cfe2 CH |
925 | while (!list_empty(&sb->s_dentry_lru)) { |
926 | dentry = list_entry(sb->s_dentry_lru.prev, | |
927 | struct dentry, d_lru); | |
928 | BUG_ON(dentry->d_sb != sb); | |
929 | ||
23044507 | 930 | if (!spin_trylock(&dentry->d_lock)) { |
19156840 | 931 | spin_unlock(&sb->s_dentry_lru_lock); |
23044507 NP |
932 | cpu_relax(); |
933 | goto relock; | |
934 | } | |
935 | ||
b48f03b3 | 936 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
937 | dentry->d_flags &= ~DCACHE_REFERENCED; |
938 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 939 | spin_unlock(&dentry->d_lock); |
23044507 | 940 | } else { |
dd1f6b2e | 941 | list_move(&dentry->d_lru, &tmp); |
eaf5f907 | 942 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
dd1f6b2e DC |
943 | this_cpu_dec(nr_dentry_unused); |
944 | sb->s_nr_dentry_unused--; | |
23044507 | 945 | spin_unlock(&dentry->d_lock); |
0a234c6d DC |
946 | freed++; |
947 | if (!--nr_to_scan) | |
23044507 | 948 | break; |
3049cfe2 | 949 | } |
19156840 | 950 | cond_resched_lock(&sb->s_dentry_lru_lock); |
3049cfe2 | 951 | } |
da3bbdd4 KM |
952 | if (!list_empty(&referenced)) |
953 | list_splice(&referenced, &sb->s_dentry_lru); | |
19156840 | 954 | spin_unlock(&sb->s_dentry_lru_lock); |
ec33679d NP |
955 | |
956 | shrink_dentry_list(&tmp); | |
0a234c6d | 957 | return freed; |
da3bbdd4 KM |
958 | } |
959 | ||
dd1f6b2e DC |
960 | /* |
961 | * Mark all the dentries as on being the dispose list so we don't think they are | |
962 | * still on the LRU if we try to kill them from ascending the parent chain in | |
963 | * try_prune_one_dentry() rather than directly from the dispose list. | |
964 | */ | |
965 | static void | |
966 | shrink_dcache_list( | |
967 | struct list_head *dispose) | |
968 | { | |
969 | struct dentry *dentry; | |
970 | ||
971 | rcu_read_lock(); | |
972 | list_for_each_entry_rcu(dentry, dispose, d_lru) { | |
973 | spin_lock(&dentry->d_lock); | |
974 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
975 | spin_unlock(&dentry->d_lock); | |
976 | } | |
977 | rcu_read_unlock(); | |
978 | shrink_dentry_list(dispose); | |
979 | } | |
980 | ||
1da177e4 LT |
981 | /** |
982 | * shrink_dcache_sb - shrink dcache for a superblock | |
983 | * @sb: superblock | |
984 | * | |
3049cfe2 CH |
985 | * Shrink the dcache for the specified super block. This is used to free |
986 | * the dcache before unmounting a file system. | |
1da177e4 | 987 | */ |
3049cfe2 | 988 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 989 | { |
3049cfe2 CH |
990 | LIST_HEAD(tmp); |
991 | ||
19156840 | 992 | spin_lock(&sb->s_dentry_lru_lock); |
3049cfe2 | 993 | while (!list_empty(&sb->s_dentry_lru)) { |
dd1f6b2e DC |
994 | /* |
995 | * account for removal here so we don't need to handle it later | |
996 | * even though the dentry is no longer on the lru list. | |
997 | */ | |
3049cfe2 | 998 | list_splice_init(&sb->s_dentry_lru, &tmp); |
dd1f6b2e DC |
999 | this_cpu_sub(nr_dentry_unused, sb->s_nr_dentry_unused); |
1000 | sb->s_nr_dentry_unused = 0; | |
19156840 | 1001 | spin_unlock(&sb->s_dentry_lru_lock); |
dd1f6b2e DC |
1002 | |
1003 | shrink_dcache_list(&tmp); | |
1004 | ||
19156840 | 1005 | spin_lock(&sb->s_dentry_lru_lock); |
3049cfe2 | 1006 | } |
19156840 | 1007 | spin_unlock(&sb->s_dentry_lru_lock); |
1da177e4 | 1008 | } |
ec4f8605 | 1009 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1010 | |
c636ebdb DH |
1011 | /* |
1012 | * destroy a single subtree of dentries for unmount | |
1013 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
1014 | * locking | |
1015 | */ | |
1016 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
1017 | { | |
1018 | struct dentry *parent; | |
1019 | ||
1020 | BUG_ON(!IS_ROOT(dentry)); | |
1021 | ||
c636ebdb DH |
1022 | for (;;) { |
1023 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 1024 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
1025 | dentry = list_entry(dentry->d_subdirs.next, |
1026 | struct dentry, d_u.d_child); | |
c636ebdb DH |
1027 | |
1028 | /* consume the dentries from this leaf up through its parents | |
1029 | * until we find one with children or run out altogether */ | |
1030 | do { | |
1031 | struct inode *inode; | |
1032 | ||
f0023bc6 | 1033 | /* |
61572bb1 | 1034 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
1035 | * unhashed and destroyed. |
1036 | */ | |
590fb51f YZ |
1037 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && |
1038 | !d_unhashed(dentry)) | |
61572bb1 YZ |
1039 | dentry->d_op->d_prune(dentry); |
1040 | ||
1041 | dentry_lru_del(dentry); | |
43c1c9cd DH |
1042 | __d_shrink(dentry); |
1043 | ||
98474236 | 1044 | if (dentry->d_lockref.count != 0) { |
c636ebdb DH |
1045 | printk(KERN_ERR |
1046 | "BUG: Dentry %p{i=%lx,n=%s}" | |
1047 | " still in use (%d)" | |
1048 | " [unmount of %s %s]\n", | |
1049 | dentry, | |
1050 | dentry->d_inode ? | |
1051 | dentry->d_inode->i_ino : 0UL, | |
1052 | dentry->d_name.name, | |
98474236 | 1053 | dentry->d_lockref.count, |
c636ebdb DH |
1054 | dentry->d_sb->s_type->name, |
1055 | dentry->d_sb->s_id); | |
1056 | BUG(); | |
1057 | } | |
1058 | ||
2fd6b7f5 | 1059 | if (IS_ROOT(dentry)) { |
c636ebdb | 1060 | parent = NULL; |
2fd6b7f5 NP |
1061 | list_del(&dentry->d_u.d_child); |
1062 | } else { | |
871c0067 | 1063 | parent = dentry->d_parent; |
98474236 | 1064 | parent->d_lockref.count--; |
2fd6b7f5 | 1065 | list_del(&dentry->d_u.d_child); |
871c0067 | 1066 | } |
c636ebdb | 1067 | |
c636ebdb DH |
1068 | inode = dentry->d_inode; |
1069 | if (inode) { | |
1070 | dentry->d_inode = NULL; | |
b3d9b7a3 | 1071 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
1072 | if (dentry->d_op && dentry->d_op->d_iput) |
1073 | dentry->d_op->d_iput(dentry, inode); | |
1074 | else | |
1075 | iput(inode); | |
1076 | } | |
1077 | ||
1078 | d_free(dentry); | |
1079 | ||
1080 | /* finished when we fall off the top of the tree, | |
1081 | * otherwise we ascend to the parent and move to the | |
1082 | * next sibling if there is one */ | |
1083 | if (!parent) | |
312d3ca8 | 1084 | return; |
c636ebdb | 1085 | dentry = parent; |
c636ebdb DH |
1086 | } while (list_empty(&dentry->d_subdirs)); |
1087 | ||
1088 | dentry = list_entry(dentry->d_subdirs.next, | |
1089 | struct dentry, d_u.d_child); | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 1095 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
1096 | * - the superblock is detached from all mountings and open files, so the |
1097 | * dentry trees will not be rearranged by the VFS | |
1098 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
1099 | * any dentries belonging to this superblock that it comes across | |
1100 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
1101 | * in this superblock | |
1102 | */ | |
1103 | void shrink_dcache_for_umount(struct super_block *sb) | |
1104 | { | |
1105 | struct dentry *dentry; | |
1106 | ||
1107 | if (down_read_trylock(&sb->s_umount)) | |
1108 | BUG(); | |
1109 | ||
1110 | dentry = sb->s_root; | |
1111 | sb->s_root = NULL; | |
98474236 | 1112 | dentry->d_lockref.count--; |
c636ebdb DH |
1113 | shrink_dcache_for_umount_subtree(dentry); |
1114 | ||
ceb5bdc2 NP |
1115 | while (!hlist_bl_empty(&sb->s_anon)) { |
1116 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1117 | shrink_dcache_for_umount_subtree(dentry); |
1118 | } | |
1119 | } | |
1120 | ||
c826cb7d LT |
1121 | /* |
1122 | * This tries to ascend one level of parenthood, but | |
1123 | * we can race with renaming, so we need to re-check | |
1124 | * the parenthood after dropping the lock and check | |
1125 | * that the sequence number still matches. | |
1126 | */ | |
48f5ec21 | 1127 | static struct dentry *try_to_ascend(struct dentry *old, unsigned seq) |
c826cb7d LT |
1128 | { |
1129 | struct dentry *new = old->d_parent; | |
1130 | ||
1131 | rcu_read_lock(); | |
1132 | spin_unlock(&old->d_lock); | |
1133 | spin_lock(&new->d_lock); | |
1134 | ||
1135 | /* | |
1136 | * might go back up the wrong parent if we have had a rename | |
1137 | * or deletion | |
1138 | */ | |
1139 | if (new != old->d_parent || | |
b161dfa6 | 1140 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
48f5ec21 | 1141 | need_seqretry(&rename_lock, seq)) { |
c826cb7d LT |
1142 | spin_unlock(&new->d_lock); |
1143 | new = NULL; | |
1144 | } | |
1145 | rcu_read_unlock(); | |
1146 | return new; | |
1147 | } | |
1148 | ||
db14fc3a MS |
1149 | /** |
1150 | * enum d_walk_ret - action to talke during tree walk | |
1151 | * @D_WALK_CONTINUE: contrinue walk | |
1152 | * @D_WALK_QUIT: quit walk | |
1153 | * @D_WALK_NORETRY: quit when retry is needed | |
1154 | * @D_WALK_SKIP: skip this dentry and its children | |
1155 | */ | |
1156 | enum d_walk_ret { | |
1157 | D_WALK_CONTINUE, | |
1158 | D_WALK_QUIT, | |
1159 | D_WALK_NORETRY, | |
1160 | D_WALK_SKIP, | |
1161 | }; | |
c826cb7d | 1162 | |
1da177e4 | 1163 | /** |
db14fc3a MS |
1164 | * d_walk - walk the dentry tree |
1165 | * @parent: start of walk | |
1166 | * @data: data passed to @enter() and @finish() | |
1167 | * @enter: callback when first entering the dentry | |
1168 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1169 | * |
db14fc3a | 1170 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1171 | */ |
db14fc3a MS |
1172 | static void d_walk(struct dentry *parent, void *data, |
1173 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1174 | void (*finish)(void *)) | |
1da177e4 | 1175 | { |
949854d0 | 1176 | struct dentry *this_parent; |
1da177e4 | 1177 | struct list_head *next; |
48f5ec21 | 1178 | unsigned seq = 0; |
db14fc3a MS |
1179 | enum d_walk_ret ret; |
1180 | bool retry = true; | |
949854d0 | 1181 | |
58db63d0 | 1182 | again: |
48f5ec21 | 1183 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1184 | this_parent = parent; |
2fd6b7f5 | 1185 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1186 | |
1187 | ret = enter(data, this_parent); | |
1188 | switch (ret) { | |
1189 | case D_WALK_CONTINUE: | |
1190 | break; | |
1191 | case D_WALK_QUIT: | |
1192 | case D_WALK_SKIP: | |
1193 | goto out_unlock; | |
1194 | case D_WALK_NORETRY: | |
1195 | retry = false; | |
1196 | break; | |
1197 | } | |
1da177e4 LT |
1198 | repeat: |
1199 | next = this_parent->d_subdirs.next; | |
1200 | resume: | |
1201 | while (next != &this_parent->d_subdirs) { | |
1202 | struct list_head *tmp = next; | |
5160ee6f | 1203 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1204 | next = tmp->next; |
2fd6b7f5 NP |
1205 | |
1206 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1207 | |
1208 | ret = enter(data, dentry); | |
1209 | switch (ret) { | |
1210 | case D_WALK_CONTINUE: | |
1211 | break; | |
1212 | case D_WALK_QUIT: | |
2fd6b7f5 | 1213 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1214 | goto out_unlock; |
1215 | case D_WALK_NORETRY: | |
1216 | retry = false; | |
1217 | break; | |
1218 | case D_WALK_SKIP: | |
1219 | spin_unlock(&dentry->d_lock); | |
1220 | continue; | |
2fd6b7f5 | 1221 | } |
db14fc3a | 1222 | |
1da177e4 | 1223 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1224 | spin_unlock(&this_parent->d_lock); |
1225 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1226 | this_parent = dentry; |
2fd6b7f5 | 1227 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1228 | goto repeat; |
1229 | } | |
2fd6b7f5 | 1230 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1231 | } |
1232 | /* | |
1233 | * All done at this level ... ascend and resume the search. | |
1234 | */ | |
1235 | if (this_parent != parent) { | |
c826cb7d | 1236 | struct dentry *child = this_parent; |
48f5ec21 | 1237 | this_parent = try_to_ascend(this_parent, seq); |
c826cb7d | 1238 | if (!this_parent) |
949854d0 | 1239 | goto rename_retry; |
949854d0 | 1240 | next = child->d_u.d_child.next; |
1da177e4 LT |
1241 | goto resume; |
1242 | } | |
48f5ec21 | 1243 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1244 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1245 | goto rename_retry; |
db14fc3a MS |
1246 | } |
1247 | if (finish) | |
1248 | finish(data); | |
1249 | ||
1250 | out_unlock: | |
1251 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1252 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1253 | return; |
58db63d0 NP |
1254 | |
1255 | rename_retry: | |
db14fc3a MS |
1256 | if (!retry) |
1257 | return; | |
48f5ec21 | 1258 | seq = 1; |
58db63d0 | 1259 | goto again; |
1da177e4 | 1260 | } |
db14fc3a MS |
1261 | |
1262 | /* | |
1263 | * Search for at least 1 mount point in the dentry's subdirs. | |
1264 | * We descend to the next level whenever the d_subdirs | |
1265 | * list is non-empty and continue searching. | |
1266 | */ | |
1267 | ||
1268 | /** | |
1269 | * have_submounts - check for mounts over a dentry | |
1270 | * @parent: dentry to check. | |
1271 | * | |
1272 | * Return true if the parent or its subdirectories contain | |
1273 | * a mount point | |
1274 | */ | |
1275 | ||
1276 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1277 | { | |
1278 | int *ret = data; | |
1279 | if (d_mountpoint(dentry)) { | |
1280 | *ret = 1; | |
1281 | return D_WALK_QUIT; | |
1282 | } | |
1283 | return D_WALK_CONTINUE; | |
1284 | } | |
1285 | ||
1286 | int have_submounts(struct dentry *parent) | |
1287 | { | |
1288 | int ret = 0; | |
1289 | ||
1290 | d_walk(parent, &ret, check_mount, NULL); | |
1291 | ||
1292 | return ret; | |
1293 | } | |
ec4f8605 | 1294 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1295 | |
eed81007 MS |
1296 | /* |
1297 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1298 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1299 | * subtree can become unreachable). | |
1300 | * | |
1301 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1302 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1303 | */ | |
1304 | int d_set_mounted(struct dentry *dentry) | |
1305 | { | |
1306 | struct dentry *p; | |
1307 | int ret = -ENOENT; | |
1308 | write_seqlock(&rename_lock); | |
1309 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1310 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1311 | spin_lock(&p->d_lock); | |
1312 | if (unlikely(d_unhashed(p))) { | |
1313 | spin_unlock(&p->d_lock); | |
1314 | goto out; | |
1315 | } | |
1316 | spin_unlock(&p->d_lock); | |
1317 | } | |
1318 | spin_lock(&dentry->d_lock); | |
1319 | if (!d_unlinked(dentry)) { | |
1320 | dentry->d_flags |= DCACHE_MOUNTED; | |
1321 | ret = 0; | |
1322 | } | |
1323 | spin_unlock(&dentry->d_lock); | |
1324 | out: | |
1325 | write_sequnlock(&rename_lock); | |
1326 | return ret; | |
1327 | } | |
1328 | ||
1da177e4 | 1329 | /* |
fd517909 | 1330 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1331 | * and move any unused dentries to the end of the unused |
1332 | * list for prune_dcache(). We descend to the next level | |
1333 | * whenever the d_subdirs list is non-empty and continue | |
1334 | * searching. | |
1335 | * | |
1336 | * It returns zero iff there are no unused children, | |
1337 | * otherwise it returns the number of children moved to | |
1338 | * the end of the unused list. This may not be the total | |
1339 | * number of unused children, because select_parent can | |
1340 | * drop the lock and return early due to latency | |
1341 | * constraints. | |
1342 | */ | |
1da177e4 | 1343 | |
db14fc3a MS |
1344 | struct select_data { |
1345 | struct dentry *start; | |
1346 | struct list_head dispose; | |
1347 | int found; | |
1348 | }; | |
23044507 | 1349 | |
db14fc3a MS |
1350 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1351 | { | |
1352 | struct select_data *data = _data; | |
1353 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1354 | |
db14fc3a MS |
1355 | if (data->start == dentry) |
1356 | goto out; | |
2fd6b7f5 | 1357 | |
1da177e4 | 1358 | /* |
db14fc3a MS |
1359 | * move only zero ref count dentries to the dispose list. |
1360 | * | |
1361 | * Those which are presently on the shrink list, being processed | |
1362 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1363 | * loop in shrink_dcache_parent() might not make any progress | |
1364 | * and loop forever. | |
1da177e4 | 1365 | */ |
db14fc3a MS |
1366 | if (dentry->d_lockref.count) { |
1367 | dentry_lru_del(dentry); | |
1368 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1369 | dentry_lru_move_list(dentry, &data->dispose); | |
1370 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
1371 | data->found++; | |
1372 | ret = D_WALK_NORETRY; | |
1da177e4 | 1373 | } |
db14fc3a MS |
1374 | /* |
1375 | * We can return to the caller if we have found some (this | |
1376 | * ensures forward progress). We'll be coming back to find | |
1377 | * the rest. | |
1378 | */ | |
1379 | if (data->found && need_resched()) | |
1380 | ret = D_WALK_QUIT; | |
1da177e4 | 1381 | out: |
db14fc3a | 1382 | return ret; |
1da177e4 LT |
1383 | } |
1384 | ||
1385 | /** | |
1386 | * shrink_dcache_parent - prune dcache | |
1387 | * @parent: parent of entries to prune | |
1388 | * | |
1389 | * Prune the dcache to remove unused children of the parent dentry. | |
1390 | */ | |
db14fc3a | 1391 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1392 | { |
db14fc3a MS |
1393 | for (;;) { |
1394 | struct select_data data; | |
1da177e4 | 1395 | |
db14fc3a MS |
1396 | INIT_LIST_HEAD(&data.dispose); |
1397 | data.start = parent; | |
1398 | data.found = 0; | |
1399 | ||
1400 | d_walk(parent, &data, select_collect, NULL); | |
1401 | if (!data.found) | |
1402 | break; | |
1403 | ||
1404 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1405 | cond_resched(); |
1406 | } | |
1da177e4 | 1407 | } |
ec4f8605 | 1408 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1409 | |
848ac114 MS |
1410 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1411 | { | |
1412 | struct select_data *data = _data; | |
1413 | ||
1414 | if (d_mountpoint(dentry)) { | |
1415 | data->found = -EBUSY; | |
1416 | return D_WALK_QUIT; | |
1417 | } | |
1418 | ||
1419 | return select_collect(_data, dentry); | |
1420 | } | |
1421 | ||
1422 | static void check_and_drop(void *_data) | |
1423 | { | |
1424 | struct select_data *data = _data; | |
1425 | ||
1426 | if (d_mountpoint(data->start)) | |
1427 | data->found = -EBUSY; | |
1428 | if (!data->found) | |
1429 | __d_drop(data->start); | |
1430 | } | |
1431 | ||
1432 | /** | |
1433 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1434 | * | |
1435 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1436 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1437 | * return -EBUSY. | |
1438 | * | |
1439 | * @dentry: dentry to prune and drop | |
1440 | */ | |
1441 | int check_submounts_and_drop(struct dentry *dentry) | |
1442 | { | |
1443 | int ret = 0; | |
1444 | ||
1445 | /* Negative dentries can be dropped without further checks */ | |
1446 | if (!dentry->d_inode) { | |
1447 | d_drop(dentry); | |
1448 | goto out; | |
1449 | } | |
1450 | ||
1451 | for (;;) { | |
1452 | struct select_data data; | |
1453 | ||
1454 | INIT_LIST_HEAD(&data.dispose); | |
1455 | data.start = dentry; | |
1456 | data.found = 0; | |
1457 | ||
1458 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1459 | ret = data.found; | |
1460 | ||
1461 | if (!list_empty(&data.dispose)) | |
1462 | shrink_dentry_list(&data.dispose); | |
1463 | ||
1464 | if (ret <= 0) | |
1465 | break; | |
1466 | ||
1467 | cond_resched(); | |
1468 | } | |
1469 | ||
1470 | out: | |
1471 | return ret; | |
1472 | } | |
1473 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1474 | ||
1da177e4 | 1475 | /** |
a4464dbc AV |
1476 | * __d_alloc - allocate a dcache entry |
1477 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1478 | * @name: qstr of the name |
1479 | * | |
1480 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1481 | * available. On a success the dentry is returned. The name passed in is | |
1482 | * copied and the copy passed in may be reused after this call. | |
1483 | */ | |
1484 | ||
a4464dbc | 1485 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1486 | { |
1487 | struct dentry *dentry; | |
1488 | char *dname; | |
1489 | ||
e12ba74d | 1490 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1491 | if (!dentry) |
1492 | return NULL; | |
1493 | ||
6326c71f LT |
1494 | /* |
1495 | * We guarantee that the inline name is always NUL-terminated. | |
1496 | * This way the memcpy() done by the name switching in rename | |
1497 | * will still always have a NUL at the end, even if we might | |
1498 | * be overwriting an internal NUL character | |
1499 | */ | |
1500 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1501 | if (name->len > DNAME_INLINE_LEN-1) { |
1502 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1503 | if (!dname) { | |
1504 | kmem_cache_free(dentry_cache, dentry); | |
1505 | return NULL; | |
1506 | } | |
1507 | } else { | |
1508 | dname = dentry->d_iname; | |
1509 | } | |
1da177e4 LT |
1510 | |
1511 | dentry->d_name.len = name->len; | |
1512 | dentry->d_name.hash = name->hash; | |
1513 | memcpy(dname, name->name, name->len); | |
1514 | dname[name->len] = 0; | |
1515 | ||
6326c71f LT |
1516 | /* Make sure we always see the terminating NUL character */ |
1517 | smp_wmb(); | |
1518 | dentry->d_name.name = dname; | |
1519 | ||
98474236 | 1520 | dentry->d_lockref.count = 1; |
dea3667b | 1521 | dentry->d_flags = 0; |
1da177e4 | 1522 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1523 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1524 | dentry->d_inode = NULL; |
a4464dbc AV |
1525 | dentry->d_parent = dentry; |
1526 | dentry->d_sb = sb; | |
1da177e4 LT |
1527 | dentry->d_op = NULL; |
1528 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1529 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1530 | INIT_LIST_HEAD(&dentry->d_lru); |
1531 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1532 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1533 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1534 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1535 | |
3e880fb5 | 1536 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1537 | |
1da177e4 LT |
1538 | return dentry; |
1539 | } | |
a4464dbc AV |
1540 | |
1541 | /** | |
1542 | * d_alloc - allocate a dcache entry | |
1543 | * @parent: parent of entry to allocate | |
1544 | * @name: qstr of the name | |
1545 | * | |
1546 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1547 | * available. On a success the dentry is returned. The name passed in is | |
1548 | * copied and the copy passed in may be reused after this call. | |
1549 | */ | |
1550 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1551 | { | |
1552 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1553 | if (!dentry) | |
1554 | return NULL; | |
1555 | ||
1556 | spin_lock(&parent->d_lock); | |
1557 | /* | |
1558 | * don't need child lock because it is not subject | |
1559 | * to concurrency here | |
1560 | */ | |
1561 | __dget_dlock(parent); | |
1562 | dentry->d_parent = parent; | |
1563 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1564 | spin_unlock(&parent->d_lock); | |
1565 | ||
1566 | return dentry; | |
1567 | } | |
ec4f8605 | 1568 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1569 | |
4b936885 NP |
1570 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1571 | { | |
a4464dbc AV |
1572 | struct dentry *dentry = __d_alloc(sb, name); |
1573 | if (dentry) | |
4b936885 | 1574 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1575 | return dentry; |
1576 | } | |
1577 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1578 | ||
1da177e4 LT |
1579 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1580 | { | |
1581 | struct qstr q; | |
1582 | ||
1583 | q.name = name; | |
1584 | q.len = strlen(name); | |
1585 | q.hash = full_name_hash(q.name, q.len); | |
1586 | return d_alloc(parent, &q); | |
1587 | } | |
ef26ca97 | 1588 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1589 | |
fb045adb NP |
1590 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1591 | { | |
6f7f7caa LT |
1592 | WARN_ON_ONCE(dentry->d_op); |
1593 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1594 | DCACHE_OP_COMPARE | |
1595 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1596 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1597 | DCACHE_OP_DELETE )); |
1598 | dentry->d_op = op; | |
1599 | if (!op) | |
1600 | return; | |
1601 | if (op->d_hash) | |
1602 | dentry->d_flags |= DCACHE_OP_HASH; | |
1603 | if (op->d_compare) | |
1604 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1605 | if (op->d_revalidate) | |
1606 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1607 | if (op->d_weak_revalidate) |
1608 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1609 | if (op->d_delete) |
1610 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1611 | if (op->d_prune) |
1612 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1613 | |
1614 | } | |
1615 | EXPORT_SYMBOL(d_set_d_op); | |
1616 | ||
360da900 OH |
1617 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1618 | { | |
b23fb0a6 | 1619 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1620 | if (inode) { |
1621 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1622 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1623 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1624 | } |
360da900 | 1625 | dentry->d_inode = inode; |
31e6b01f | 1626 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1627 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1628 | fsnotify_d_instantiate(dentry, inode); |
1629 | } | |
1630 | ||
1da177e4 LT |
1631 | /** |
1632 | * d_instantiate - fill in inode information for a dentry | |
1633 | * @entry: dentry to complete | |
1634 | * @inode: inode to attach to this dentry | |
1635 | * | |
1636 | * Fill in inode information in the entry. | |
1637 | * | |
1638 | * This turns negative dentries into productive full members | |
1639 | * of society. | |
1640 | * | |
1641 | * NOTE! This assumes that the inode count has been incremented | |
1642 | * (or otherwise set) by the caller to indicate that it is now | |
1643 | * in use by the dcache. | |
1644 | */ | |
1645 | ||
1646 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1647 | { | |
b3d9b7a3 | 1648 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1649 | if (inode) |
1650 | spin_lock(&inode->i_lock); | |
360da900 | 1651 | __d_instantiate(entry, inode); |
873feea0 NP |
1652 | if (inode) |
1653 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1654 | security_d_instantiate(entry, inode); |
1655 | } | |
ec4f8605 | 1656 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1657 | |
1658 | /** | |
1659 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1660 | * @entry: dentry to instantiate | |
1661 | * @inode: inode to attach to this dentry | |
1662 | * | |
1663 | * Fill in inode information in the entry. On success, it returns NULL. | |
1664 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1665 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1666 | * |
1667 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1668 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1669 | * |
1670 | * This also assumes that the inode count has been incremented | |
1671 | * (or otherwise set) by the caller to indicate that it is now | |
1672 | * in use by the dcache. | |
1da177e4 | 1673 | */ |
770bfad8 DH |
1674 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1675 | struct inode *inode) | |
1da177e4 LT |
1676 | { |
1677 | struct dentry *alias; | |
1678 | int len = entry->d_name.len; | |
1679 | const char *name = entry->d_name.name; | |
1680 | unsigned int hash = entry->d_name.hash; | |
1681 | ||
770bfad8 | 1682 | if (!inode) { |
360da900 | 1683 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1684 | return NULL; |
1685 | } | |
1686 | ||
b67bfe0d | 1687 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1688 | /* |
1689 | * Don't need alias->d_lock here, because aliases with | |
1690 | * d_parent == entry->d_parent are not subject to name or | |
1691 | * parent changes, because the parent inode i_mutex is held. | |
1692 | */ | |
12f8ad4b | 1693 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1694 | continue; |
1695 | if (alias->d_parent != entry->d_parent) | |
1696 | continue; | |
ee983e89 LT |
1697 | if (alias->d_name.len != len) |
1698 | continue; | |
12f8ad4b | 1699 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1700 | continue; |
dc0474be | 1701 | __dget(alias); |
1da177e4 LT |
1702 | return alias; |
1703 | } | |
770bfad8 | 1704 | |
360da900 | 1705 | __d_instantiate(entry, inode); |
1da177e4 LT |
1706 | return NULL; |
1707 | } | |
770bfad8 DH |
1708 | |
1709 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1710 | { | |
1711 | struct dentry *result; | |
1712 | ||
b3d9b7a3 | 1713 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1714 | |
873feea0 NP |
1715 | if (inode) |
1716 | spin_lock(&inode->i_lock); | |
770bfad8 | 1717 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1718 | if (inode) |
1719 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1720 | |
1721 | if (!result) { | |
1722 | security_d_instantiate(entry, inode); | |
1723 | return NULL; | |
1724 | } | |
1725 | ||
1726 | BUG_ON(!d_unhashed(result)); | |
1727 | iput(inode); | |
1728 | return result; | |
1729 | } | |
1730 | ||
1da177e4 LT |
1731 | EXPORT_SYMBOL(d_instantiate_unique); |
1732 | ||
adc0e91a AV |
1733 | struct dentry *d_make_root(struct inode *root_inode) |
1734 | { | |
1735 | struct dentry *res = NULL; | |
1736 | ||
1737 | if (root_inode) { | |
26fe5750 | 1738 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1739 | |
1740 | res = __d_alloc(root_inode->i_sb, &name); | |
1741 | if (res) | |
1742 | d_instantiate(res, root_inode); | |
1743 | else | |
1744 | iput(root_inode); | |
1745 | } | |
1746 | return res; | |
1747 | } | |
1748 | EXPORT_SYMBOL(d_make_root); | |
1749 | ||
d891eedb BF |
1750 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1751 | { | |
1752 | struct dentry *alias; | |
1753 | ||
b3d9b7a3 | 1754 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1755 | return NULL; |
b3d9b7a3 | 1756 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1757 | __dget(alias); |
1758 | return alias; | |
1759 | } | |
1760 | ||
46f72b34 SW |
1761 | /** |
1762 | * d_find_any_alias - find any alias for a given inode | |
1763 | * @inode: inode to find an alias for | |
1764 | * | |
1765 | * If any aliases exist for the given inode, take and return a | |
1766 | * reference for one of them. If no aliases exist, return %NULL. | |
1767 | */ | |
1768 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1769 | { |
1770 | struct dentry *de; | |
1771 | ||
1772 | spin_lock(&inode->i_lock); | |
1773 | de = __d_find_any_alias(inode); | |
1774 | spin_unlock(&inode->i_lock); | |
1775 | return de; | |
1776 | } | |
46f72b34 | 1777 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1778 | |
4ea3ada2 CH |
1779 | /** |
1780 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1781 | * @inode: inode to allocate the dentry for | |
1782 | * | |
1783 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1784 | * similar open by handle operations. The returned dentry may be anonymous, | |
1785 | * or may have a full name (if the inode was already in the cache). | |
1786 | * | |
1787 | * When called on a directory inode, we must ensure that the inode only ever | |
1788 | * has one dentry. If a dentry is found, that is returned instead of | |
1789 | * allocating a new one. | |
1790 | * | |
1791 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1792 | * to the dentry. In case of an error the reference on the inode is released. |
1793 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1794 | * be passed in and will be the error will be propagate to the return value, | |
1795 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1796 | */ |
1797 | struct dentry *d_obtain_alias(struct inode *inode) | |
1798 | { | |
b911a6bd | 1799 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1800 | struct dentry *tmp; |
1801 | struct dentry *res; | |
4ea3ada2 CH |
1802 | |
1803 | if (!inode) | |
44003728 | 1804 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1805 | if (IS_ERR(inode)) |
1806 | return ERR_CAST(inode); | |
1807 | ||
d891eedb | 1808 | res = d_find_any_alias(inode); |
9308a612 CH |
1809 | if (res) |
1810 | goto out_iput; | |
1811 | ||
a4464dbc | 1812 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1813 | if (!tmp) { |
1814 | res = ERR_PTR(-ENOMEM); | |
1815 | goto out_iput; | |
4ea3ada2 | 1816 | } |
b5c84bf6 | 1817 | |
873feea0 | 1818 | spin_lock(&inode->i_lock); |
d891eedb | 1819 | res = __d_find_any_alias(inode); |
9308a612 | 1820 | if (res) { |
873feea0 | 1821 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1822 | dput(tmp); |
1823 | goto out_iput; | |
1824 | } | |
1825 | ||
1826 | /* attach a disconnected dentry */ | |
1827 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1828 | tmp->d_inode = inode; |
1829 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1830 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1831 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1832 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1833 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1834 | spin_unlock(&tmp->d_lock); |
873feea0 | 1835 | spin_unlock(&inode->i_lock); |
24ff6663 | 1836 | security_d_instantiate(tmp, inode); |
9308a612 | 1837 | |
9308a612 CH |
1838 | return tmp; |
1839 | ||
1840 | out_iput: | |
24ff6663 JB |
1841 | if (res && !IS_ERR(res)) |
1842 | security_d_instantiate(res, inode); | |
9308a612 CH |
1843 | iput(inode); |
1844 | return res; | |
4ea3ada2 | 1845 | } |
adc48720 | 1846 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1847 | |
1848 | /** | |
1849 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1850 | * @inode: the inode which may have a disconnected dentry | |
1851 | * @dentry: a negative dentry which we want to point to the inode. | |
1852 | * | |
1853 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1854 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1855 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1856 | * | |
1857 | * This is needed in the lookup routine of any filesystem that is exportable | |
1858 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1859 | * | |
1860 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1861 | * is returned. This matches the expected return value of ->lookup. | |
1862 | * | |
6d4ade98 SW |
1863 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1864 | * In that case, we know that the inode will be a regular file, and also this | |
1865 | * will only occur during atomic_open. So we need to check for the dentry | |
1866 | * being already hashed only in the final case. | |
1da177e4 LT |
1867 | */ |
1868 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1869 | { | |
1870 | struct dentry *new = NULL; | |
1871 | ||
a9049376 AV |
1872 | if (IS_ERR(inode)) |
1873 | return ERR_CAST(inode); | |
1874 | ||
21c0d8fd | 1875 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1876 | spin_lock(&inode->i_lock); |
32ba9c3f | 1877 | new = __d_find_alias(inode, 1); |
1da177e4 | 1878 | if (new) { |
32ba9c3f | 1879 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1880 | spin_unlock(&inode->i_lock); |
1da177e4 | 1881 | security_d_instantiate(new, inode); |
1da177e4 LT |
1882 | d_move(new, dentry); |
1883 | iput(inode); | |
1884 | } else { | |
873feea0 | 1885 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1886 | __d_instantiate(dentry, inode); |
873feea0 | 1887 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1888 | security_d_instantiate(dentry, inode); |
1889 | d_rehash(dentry); | |
1890 | } | |
6d4ade98 SW |
1891 | } else { |
1892 | d_instantiate(dentry, inode); | |
1893 | if (d_unhashed(dentry)) | |
1894 | d_rehash(dentry); | |
1895 | } | |
1da177e4 LT |
1896 | return new; |
1897 | } | |
ec4f8605 | 1898 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1899 | |
9403540c BN |
1900 | /** |
1901 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1902 | * @inode: the inode case-insensitive lookup has found | |
1903 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1904 | * @name: the case-exact name to be associated with the returned dentry | |
1905 | * | |
1906 | * This is to avoid filling the dcache with case-insensitive names to the | |
1907 | * same inode, only the actual correct case is stored in the dcache for | |
1908 | * case-insensitive filesystems. | |
1909 | * | |
1910 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1911 | * already exists in in the dcache, use it and return it. | |
1912 | * | |
1913 | * If no entry exists with the exact case name, allocate new dentry with | |
1914 | * the exact case, and return the spliced entry. | |
1915 | */ | |
e45b590b | 1916 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1917 | struct qstr *name) |
1918 | { | |
9403540c BN |
1919 | struct dentry *found; |
1920 | struct dentry *new; | |
1921 | ||
b6520c81 CH |
1922 | /* |
1923 | * First check if a dentry matching the name already exists, | |
1924 | * if not go ahead and create it now. | |
1925 | */ | |
9403540c | 1926 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1927 | if (unlikely(IS_ERR(found))) |
1928 | goto err_out; | |
9403540c BN |
1929 | if (!found) { |
1930 | new = d_alloc(dentry->d_parent, name); | |
1931 | if (!new) { | |
4f522a24 | 1932 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1933 | goto err_out; |
1934 | } | |
b6520c81 | 1935 | |
9403540c BN |
1936 | found = d_splice_alias(inode, new); |
1937 | if (found) { | |
1938 | dput(new); | |
1939 | return found; | |
1940 | } | |
1941 | return new; | |
1942 | } | |
b6520c81 CH |
1943 | |
1944 | /* | |
1945 | * If a matching dentry exists, and it's not negative use it. | |
1946 | * | |
1947 | * Decrement the reference count to balance the iget() done | |
1948 | * earlier on. | |
1949 | */ | |
9403540c BN |
1950 | if (found->d_inode) { |
1951 | if (unlikely(found->d_inode != inode)) { | |
1952 | /* This can't happen because bad inodes are unhashed. */ | |
1953 | BUG_ON(!is_bad_inode(inode)); | |
1954 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1955 | } | |
9403540c BN |
1956 | iput(inode); |
1957 | return found; | |
1958 | } | |
b6520c81 | 1959 | |
9403540c | 1960 | /* |
9403540c | 1961 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1962 | * already has a dentry. |
9403540c | 1963 | */ |
4513d899 AV |
1964 | new = d_splice_alias(inode, found); |
1965 | if (new) { | |
1966 | dput(found); | |
1967 | found = new; | |
9403540c | 1968 | } |
4513d899 | 1969 | return found; |
9403540c BN |
1970 | |
1971 | err_out: | |
1972 | iput(inode); | |
4f522a24 | 1973 | return found; |
9403540c | 1974 | } |
ec4f8605 | 1975 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1976 | |
12f8ad4b LT |
1977 | /* |
1978 | * Do the slow-case of the dentry name compare. | |
1979 | * | |
1980 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1981 | * load the name and length information, so that the |
12f8ad4b LT |
1982 | * filesystem can rely on them, and can use the 'name' and |
1983 | * 'len' information without worrying about walking off the | |
1984 | * end of memory etc. | |
1985 | * | |
1986 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1987 | * in arguments (the low-level filesystem should not look | |
1988 | * at the dentry inode or name contents directly, since | |
1989 | * rename can change them while we're in RCU mode). | |
1990 | */ | |
1991 | enum slow_d_compare { | |
1992 | D_COMP_OK, | |
1993 | D_COMP_NOMATCH, | |
1994 | D_COMP_SEQRETRY, | |
1995 | }; | |
1996 | ||
1997 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1998 | const struct dentry *parent, | |
12f8ad4b LT |
1999 | struct dentry *dentry, |
2000 | unsigned int seq, | |
2001 | const struct qstr *name) | |
2002 | { | |
2003 | int tlen = dentry->d_name.len; | |
2004 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2005 | |
2006 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2007 | cpu_relax(); | |
2008 | return D_COMP_SEQRETRY; | |
2009 | } | |
da53be12 | 2010 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2011 | return D_COMP_NOMATCH; |
2012 | return D_COMP_OK; | |
2013 | } | |
2014 | ||
31e6b01f NP |
2015 | /** |
2016 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2017 | * @parent: parent dentry | |
2018 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2019 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2020 | * Returns: dentry, or NULL |
2021 | * | |
2022 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2023 | * resolution (store-free path walking) design described in | |
2024 | * Documentation/filesystems/path-lookup.txt. | |
2025 | * | |
2026 | * This is not to be used outside core vfs. | |
2027 | * | |
2028 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2029 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2030 | * without taking d_lock and checking d_seq sequence count against @seq | |
2031 | * returned here. | |
2032 | * | |
15570086 | 2033 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2034 | * function. |
2035 | * | |
2036 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2037 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2038 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2039 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2040 | * |
2041 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2042 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2043 | */ |
8966be90 LT |
2044 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2045 | const struct qstr *name, | |
da53be12 | 2046 | unsigned *seqp) |
31e6b01f | 2047 | { |
26fe5750 | 2048 | u64 hashlen = name->hash_len; |
31e6b01f | 2049 | const unsigned char *str = name->name; |
26fe5750 | 2050 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2051 | struct hlist_bl_node *node; |
31e6b01f NP |
2052 | struct dentry *dentry; |
2053 | ||
2054 | /* | |
2055 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2056 | * required to prevent single threaded performance regressions | |
2057 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2058 | * Keep the two functions in sync. | |
2059 | */ | |
2060 | ||
2061 | /* | |
2062 | * The hash list is protected using RCU. | |
2063 | * | |
2064 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2065 | * races with d_move(). | |
2066 | * | |
2067 | * It is possible that concurrent renames can mess up our list | |
2068 | * walk here and result in missing our dentry, resulting in the | |
2069 | * false-negative result. d_lookup() protects against concurrent | |
2070 | * renames using rename_lock seqlock. | |
2071 | * | |
b0a4bb83 | 2072 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2073 | */ |
b07ad996 | 2074 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2075 | unsigned seq; |
31e6b01f | 2076 | |
31e6b01f | 2077 | seqretry: |
12f8ad4b LT |
2078 | /* |
2079 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2080 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2081 | * |
2082 | * The caller must perform a seqcount check in order | |
da53be12 | 2083 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2084 | * |
2085 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2086 | * we don't wait for the sequence count to stabilize if it | |
2087 | * is in the middle of a sequence change. If we do the slow | |
2088 | * dentry compare, we will do seqretries until it is stable, | |
2089 | * and if we end up with a successful lookup, we actually | |
2090 | * want to exit RCU lookup anyway. | |
2091 | */ | |
2092 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2093 | if (dentry->d_parent != parent) |
2094 | continue; | |
2e321806 LT |
2095 | if (d_unhashed(dentry)) |
2096 | continue; | |
12f8ad4b | 2097 | |
830c0f0e | 2098 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2099 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2100 | continue; | |
da53be12 LT |
2101 | *seqp = seq; |
2102 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2103 | case D_COMP_OK: |
2104 | return dentry; | |
2105 | case D_COMP_NOMATCH: | |
31e6b01f | 2106 | continue; |
12f8ad4b LT |
2107 | default: |
2108 | goto seqretry; | |
2109 | } | |
31e6b01f | 2110 | } |
12f8ad4b | 2111 | |
26fe5750 | 2112 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2113 | continue; |
da53be12 | 2114 | *seqp = seq; |
26fe5750 | 2115 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2116 | return dentry; |
31e6b01f NP |
2117 | } |
2118 | return NULL; | |
2119 | } | |
2120 | ||
1da177e4 LT |
2121 | /** |
2122 | * d_lookup - search for a dentry | |
2123 | * @parent: parent dentry | |
2124 | * @name: qstr of name we wish to find | |
b04f784e | 2125 | * Returns: dentry, or NULL |
1da177e4 | 2126 | * |
b04f784e NP |
2127 | * d_lookup searches the children of the parent dentry for the name in |
2128 | * question. If the dentry is found its reference count is incremented and the | |
2129 | * dentry is returned. The caller must use dput to free the entry when it has | |
2130 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2131 | */ |
da2d8455 | 2132 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2133 | { |
31e6b01f | 2134 | struct dentry *dentry; |
949854d0 | 2135 | unsigned seq; |
1da177e4 LT |
2136 | |
2137 | do { | |
2138 | seq = read_seqbegin(&rename_lock); | |
2139 | dentry = __d_lookup(parent, name); | |
2140 | if (dentry) | |
2141 | break; | |
2142 | } while (read_seqretry(&rename_lock, seq)); | |
2143 | return dentry; | |
2144 | } | |
ec4f8605 | 2145 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2146 | |
31e6b01f | 2147 | /** |
b04f784e NP |
2148 | * __d_lookup - search for a dentry (racy) |
2149 | * @parent: parent dentry | |
2150 | * @name: qstr of name we wish to find | |
2151 | * Returns: dentry, or NULL | |
2152 | * | |
2153 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2154 | * false-negative result due to unrelated rename activity. | |
2155 | * | |
2156 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2157 | * however it must be used carefully, eg. with a following d_lookup in | |
2158 | * the case of failure. | |
2159 | * | |
2160 | * __d_lookup callers must be commented. | |
2161 | */ | |
a713ca2a | 2162 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2163 | { |
2164 | unsigned int len = name->len; | |
2165 | unsigned int hash = name->hash; | |
2166 | const unsigned char *str = name->name; | |
b07ad996 | 2167 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2168 | struct hlist_bl_node *node; |
31e6b01f | 2169 | struct dentry *found = NULL; |
665a7583 | 2170 | struct dentry *dentry; |
1da177e4 | 2171 | |
31e6b01f NP |
2172 | /* |
2173 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2174 | * required to prevent single threaded performance regressions | |
2175 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2176 | * Keep the two functions in sync. | |
2177 | */ | |
2178 | ||
b04f784e NP |
2179 | /* |
2180 | * The hash list is protected using RCU. | |
2181 | * | |
2182 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2183 | * with d_move(). | |
2184 | * | |
2185 | * It is possible that concurrent renames can mess up our list | |
2186 | * walk here and result in missing our dentry, resulting in the | |
2187 | * false-negative result. d_lookup() protects against concurrent | |
2188 | * renames using rename_lock seqlock. | |
2189 | * | |
b0a4bb83 | 2190 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2191 | */ |
1da177e4 LT |
2192 | rcu_read_lock(); |
2193 | ||
b07ad996 | 2194 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2195 | |
1da177e4 LT |
2196 | if (dentry->d_name.hash != hash) |
2197 | continue; | |
1da177e4 LT |
2198 | |
2199 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2200 | if (dentry->d_parent != parent) |
2201 | goto next; | |
d0185c08 LT |
2202 | if (d_unhashed(dentry)) |
2203 | goto next; | |
2204 | ||
1da177e4 LT |
2205 | /* |
2206 | * It is safe to compare names since d_move() cannot | |
2207 | * change the qstr (protected by d_lock). | |
2208 | */ | |
fb045adb | 2209 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2210 | int tlen = dentry->d_name.len; |
2211 | const char *tname = dentry->d_name.name; | |
da53be12 | 2212 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2213 | goto next; |
2214 | } else { | |
ee983e89 LT |
2215 | if (dentry->d_name.len != len) |
2216 | goto next; | |
12f8ad4b | 2217 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2218 | goto next; |
2219 | } | |
2220 | ||
98474236 | 2221 | dentry->d_lockref.count++; |
d0185c08 | 2222 | found = dentry; |
1da177e4 LT |
2223 | spin_unlock(&dentry->d_lock); |
2224 | break; | |
2225 | next: | |
2226 | spin_unlock(&dentry->d_lock); | |
2227 | } | |
2228 | rcu_read_unlock(); | |
2229 | ||
2230 | return found; | |
2231 | } | |
2232 | ||
3e7e241f EB |
2233 | /** |
2234 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2235 | * @dir: Directory to search in | |
2236 | * @name: qstr of name we wish to find | |
2237 | * | |
4f522a24 | 2238 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2239 | */ |
2240 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2241 | { | |
3e7e241f EB |
2242 | /* |
2243 | * Check for a fs-specific hash function. Note that we must | |
2244 | * calculate the standard hash first, as the d_op->d_hash() | |
2245 | * routine may choose to leave the hash value unchanged. | |
2246 | */ | |
2247 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2248 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2249 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2250 | if (unlikely(err < 0)) |
2251 | return ERR_PTR(err); | |
3e7e241f | 2252 | } |
4f522a24 | 2253 | return d_lookup(dir, name); |
3e7e241f | 2254 | } |
4f522a24 | 2255 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2256 | |
1da177e4 | 2257 | /** |
786a5e15 | 2258 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2259 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2260 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2261 | * |
2262 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2263 | * This is used by ncpfs in its readdir implementation. | |
2264 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2265 | * |
2266 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2267 | */ |
d3a23e16 | 2268 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2269 | { |
786a5e15 | 2270 | struct dentry *child; |
d3a23e16 | 2271 | |
2fd6b7f5 | 2272 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2273 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2274 | if (dentry == child) { | |
2fd6b7f5 | 2275 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2276 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2277 | spin_unlock(&dentry->d_lock); |
2278 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2279 | return 1; |
2280 | } | |
2281 | } | |
2fd6b7f5 | 2282 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2283 | |
1da177e4 LT |
2284 | return 0; |
2285 | } | |
ec4f8605 | 2286 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2287 | |
2288 | /* | |
2289 | * When a file is deleted, we have two options: | |
2290 | * - turn this dentry into a negative dentry | |
2291 | * - unhash this dentry and free it. | |
2292 | * | |
2293 | * Usually, we want to just turn this into | |
2294 | * a negative dentry, but if anybody else is | |
2295 | * currently using the dentry or the inode | |
2296 | * we can't do that and we fall back on removing | |
2297 | * it from the hash queues and waiting for | |
2298 | * it to be deleted later when it has no users | |
2299 | */ | |
2300 | ||
2301 | /** | |
2302 | * d_delete - delete a dentry | |
2303 | * @dentry: The dentry to delete | |
2304 | * | |
2305 | * Turn the dentry into a negative dentry if possible, otherwise | |
2306 | * remove it from the hash queues so it can be deleted later | |
2307 | */ | |
2308 | ||
2309 | void d_delete(struct dentry * dentry) | |
2310 | { | |
873feea0 | 2311 | struct inode *inode; |
7a91bf7f | 2312 | int isdir = 0; |
1da177e4 LT |
2313 | /* |
2314 | * Are we the only user? | |
2315 | */ | |
357f8e65 | 2316 | again: |
1da177e4 | 2317 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2318 | inode = dentry->d_inode; |
2319 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2320 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2321 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2322 | spin_unlock(&dentry->d_lock); |
2323 | cpu_relax(); | |
2324 | goto again; | |
2325 | } | |
13e3c5e5 | 2326 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2327 | dentry_unlink_inode(dentry); |
7a91bf7f | 2328 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2329 | return; |
2330 | } | |
2331 | ||
2332 | if (!d_unhashed(dentry)) | |
2333 | __d_drop(dentry); | |
2334 | ||
2335 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2336 | |
2337 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2338 | } |
ec4f8605 | 2339 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2340 | |
b07ad996 | 2341 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2342 | { |
ceb5bdc2 | 2343 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2344 | hlist_bl_lock(b); |
dea3667b | 2345 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2346 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2347 | hlist_bl_unlock(b); |
1da177e4 LT |
2348 | } |
2349 | ||
770bfad8 DH |
2350 | static void _d_rehash(struct dentry * entry) |
2351 | { | |
2352 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2353 | } | |
2354 | ||
1da177e4 LT |
2355 | /** |
2356 | * d_rehash - add an entry back to the hash | |
2357 | * @entry: dentry to add to the hash | |
2358 | * | |
2359 | * Adds a dentry to the hash according to its name. | |
2360 | */ | |
2361 | ||
2362 | void d_rehash(struct dentry * entry) | |
2363 | { | |
1da177e4 | 2364 | spin_lock(&entry->d_lock); |
770bfad8 | 2365 | _d_rehash(entry); |
1da177e4 | 2366 | spin_unlock(&entry->d_lock); |
1da177e4 | 2367 | } |
ec4f8605 | 2368 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2369 | |
fb2d5b86 NP |
2370 | /** |
2371 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2372 | * @dentry: dentry to be updated | |
2373 | * @name: new name | |
2374 | * | |
2375 | * Update a case insensitive dentry with new case of name. | |
2376 | * | |
2377 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2378 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2379 | * lengths). | |
2380 | * | |
2381 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2382 | * keep renames and concurrent inserts, and readdir(2) away). | |
2383 | */ | |
2384 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2385 | { | |
7ebfa57f | 2386 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2387 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2388 | ||
fb2d5b86 | 2389 | spin_lock(&dentry->d_lock); |
31e6b01f | 2390 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2391 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2392 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2393 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2394 | } |
2395 | EXPORT_SYMBOL(dentry_update_name_case); | |
2396 | ||
1da177e4 LT |
2397 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2398 | { | |
2399 | if (dname_external(target)) { | |
2400 | if (dname_external(dentry)) { | |
2401 | /* | |
2402 | * Both external: swap the pointers | |
2403 | */ | |
9a8d5bb4 | 2404 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2405 | } else { |
2406 | /* | |
2407 | * dentry:internal, target:external. Steal target's | |
2408 | * storage and make target internal. | |
2409 | */ | |
321bcf92 BF |
2410 | memcpy(target->d_iname, dentry->d_name.name, |
2411 | dentry->d_name.len + 1); | |
1da177e4 LT |
2412 | dentry->d_name.name = target->d_name.name; |
2413 | target->d_name.name = target->d_iname; | |
2414 | } | |
2415 | } else { | |
2416 | if (dname_external(dentry)) { | |
2417 | /* | |
2418 | * dentry:external, target:internal. Give dentry's | |
2419 | * storage to target and make dentry internal | |
2420 | */ | |
2421 | memcpy(dentry->d_iname, target->d_name.name, | |
2422 | target->d_name.len + 1); | |
2423 | target->d_name.name = dentry->d_name.name; | |
2424 | dentry->d_name.name = dentry->d_iname; | |
2425 | } else { | |
2426 | /* | |
2427 | * Both are internal. Just copy target to dentry | |
2428 | */ | |
2429 | memcpy(dentry->d_iname, target->d_name.name, | |
2430 | target->d_name.len + 1); | |
dc711ca3 AV |
2431 | dentry->d_name.len = target->d_name.len; |
2432 | return; | |
1da177e4 LT |
2433 | } |
2434 | } | |
9a8d5bb4 | 2435 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2436 | } |
2437 | ||
2fd6b7f5 NP |
2438 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2439 | { | |
2440 | /* | |
2441 | * XXXX: do we really need to take target->d_lock? | |
2442 | */ | |
2443 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2444 | spin_lock(&target->d_parent->d_lock); | |
2445 | else { | |
2446 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2447 | spin_lock(&dentry->d_parent->d_lock); | |
2448 | spin_lock_nested(&target->d_parent->d_lock, | |
2449 | DENTRY_D_LOCK_NESTED); | |
2450 | } else { | |
2451 | spin_lock(&target->d_parent->d_lock); | |
2452 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2453 | DENTRY_D_LOCK_NESTED); | |
2454 | } | |
2455 | } | |
2456 | if (target < dentry) { | |
2457 | spin_lock_nested(&target->d_lock, 2); | |
2458 | spin_lock_nested(&dentry->d_lock, 3); | |
2459 | } else { | |
2460 | spin_lock_nested(&dentry->d_lock, 2); | |
2461 | spin_lock_nested(&target->d_lock, 3); | |
2462 | } | |
2463 | } | |
2464 | ||
2465 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2466 | struct dentry *target) | |
2467 | { | |
2468 | if (target->d_parent != dentry->d_parent) | |
2469 | spin_unlock(&dentry->d_parent->d_lock); | |
2470 | if (target->d_parent != target) | |
2471 | spin_unlock(&target->d_parent->d_lock); | |
2472 | } | |
2473 | ||
1da177e4 | 2474 | /* |
2fd6b7f5 NP |
2475 | * When switching names, the actual string doesn't strictly have to |
2476 | * be preserved in the target - because we're dropping the target | |
2477 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2478 | * the new name before we switch. | |
2479 | * | |
2480 | * Note that we have to be a lot more careful about getting the hash | |
2481 | * switched - we have to switch the hash value properly even if it | |
2482 | * then no longer matches the actual (corrupted) string of the target. | |
2483 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2484 | */ |
9eaef27b | 2485 | /* |
18367501 | 2486 | * __d_move - move a dentry |
1da177e4 LT |
2487 | * @dentry: entry to move |
2488 | * @target: new dentry | |
2489 | * | |
2490 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2491 | * dcache entries should not be moved in this way. Caller must hold |
2492 | * rename_lock, the i_mutex of the source and target directories, | |
2493 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2494 | */ |
18367501 | 2495 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2496 | { |
1da177e4 LT |
2497 | if (!dentry->d_inode) |
2498 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2499 | ||
2fd6b7f5 NP |
2500 | BUG_ON(d_ancestor(dentry, target)); |
2501 | BUG_ON(d_ancestor(target, dentry)); | |
2502 | ||
2fd6b7f5 | 2503 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2504 | |
31e6b01f NP |
2505 | write_seqcount_begin(&dentry->d_seq); |
2506 | write_seqcount_begin(&target->d_seq); | |
2507 | ||
ceb5bdc2 NP |
2508 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2509 | ||
2510 | /* | |
2511 | * Move the dentry to the target hash queue. Don't bother checking | |
2512 | * for the same hash queue because of how unlikely it is. | |
2513 | */ | |
2514 | __d_drop(dentry); | |
789680d1 | 2515 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2516 | |
2517 | /* Unhash the target: dput() will then get rid of it */ | |
2518 | __d_drop(target); | |
2519 | ||
5160ee6f ED |
2520 | list_del(&dentry->d_u.d_child); |
2521 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2522 | |
2523 | /* Switch the names.. */ | |
2524 | switch_names(dentry, target); | |
9a8d5bb4 | 2525 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2526 | |
2527 | /* ... and switch the parents */ | |
2528 | if (IS_ROOT(dentry)) { | |
2529 | dentry->d_parent = target->d_parent; | |
2530 | target->d_parent = target; | |
5160ee6f | 2531 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2532 | } else { |
9a8d5bb4 | 2533 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2534 | |
2535 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2536 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2537 | } |
2538 | ||
5160ee6f | 2539 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2540 | |
31e6b01f NP |
2541 | write_seqcount_end(&target->d_seq); |
2542 | write_seqcount_end(&dentry->d_seq); | |
2543 | ||
2fd6b7f5 | 2544 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2545 | spin_unlock(&target->d_lock); |
c32ccd87 | 2546 | fsnotify_d_move(dentry); |
1da177e4 | 2547 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2548 | } |
2549 | ||
2550 | /* | |
2551 | * d_move - move a dentry | |
2552 | * @dentry: entry to move | |
2553 | * @target: new dentry | |
2554 | * | |
2555 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2556 | * dcache entries should not be moved in this way. See the locking |
2557 | * requirements for __d_move. | |
18367501 AV |
2558 | */ |
2559 | void d_move(struct dentry *dentry, struct dentry *target) | |
2560 | { | |
2561 | write_seqlock(&rename_lock); | |
2562 | __d_move(dentry, target); | |
1da177e4 | 2563 | write_sequnlock(&rename_lock); |
9eaef27b | 2564 | } |
ec4f8605 | 2565 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2566 | |
e2761a11 OH |
2567 | /** |
2568 | * d_ancestor - search for an ancestor | |
2569 | * @p1: ancestor dentry | |
2570 | * @p2: child dentry | |
2571 | * | |
2572 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2573 | * an ancestor of p2, else NULL. | |
9eaef27b | 2574 | */ |
e2761a11 | 2575 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2576 | { |
2577 | struct dentry *p; | |
2578 | ||
871c0067 | 2579 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2580 | if (p->d_parent == p1) |
e2761a11 | 2581 | return p; |
9eaef27b | 2582 | } |
e2761a11 | 2583 | return NULL; |
9eaef27b TM |
2584 | } |
2585 | ||
2586 | /* | |
2587 | * This helper attempts to cope with remotely renamed directories | |
2588 | * | |
2589 | * It assumes that the caller is already holding | |
18367501 | 2590 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2591 | * |
2592 | * Note: If ever the locking in lock_rename() changes, then please | |
2593 | * remember to update this too... | |
9eaef27b | 2594 | */ |
873feea0 NP |
2595 | static struct dentry *__d_unalias(struct inode *inode, |
2596 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2597 | { |
2598 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2599 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2600 | |
2601 | /* If alias and dentry share a parent, then no extra locks required */ | |
2602 | if (alias->d_parent == dentry->d_parent) | |
2603 | goto out_unalias; | |
2604 | ||
9eaef27b | 2605 | /* See lock_rename() */ |
9eaef27b TM |
2606 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2607 | goto out_err; | |
2608 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2609 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2610 | goto out_err; | |
2611 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2612 | out_unalias: | |
ee3efa91 AV |
2613 | if (likely(!d_mountpoint(alias))) { |
2614 | __d_move(alias, dentry); | |
2615 | ret = alias; | |
2616 | } | |
9eaef27b | 2617 | out_err: |
873feea0 | 2618 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2619 | if (m2) |
2620 | mutex_unlock(m2); | |
2621 | if (m1) | |
2622 | mutex_unlock(m1); | |
2623 | return ret; | |
2624 | } | |
2625 | ||
770bfad8 DH |
2626 | /* |
2627 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2628 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2629 | * returns with anon->d_lock held! |
770bfad8 DH |
2630 | */ |
2631 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2632 | { | |
740da42e | 2633 | struct dentry *dparent; |
770bfad8 | 2634 | |
2fd6b7f5 | 2635 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2636 | |
31e6b01f NP |
2637 | write_seqcount_begin(&dentry->d_seq); |
2638 | write_seqcount_begin(&anon->d_seq); | |
2639 | ||
770bfad8 | 2640 | dparent = dentry->d_parent; |
770bfad8 | 2641 | |
2fd6b7f5 NP |
2642 | switch_names(dentry, anon); |
2643 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2644 | ||
740da42e AV |
2645 | dentry->d_parent = dentry; |
2646 | list_del_init(&dentry->d_u.d_child); | |
2647 | anon->d_parent = dparent; | |
9ed53b12 | 2648 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2649 | |
31e6b01f NP |
2650 | write_seqcount_end(&dentry->d_seq); |
2651 | write_seqcount_end(&anon->d_seq); | |
2652 | ||
2fd6b7f5 NP |
2653 | dentry_unlock_parents_for_move(anon, dentry); |
2654 | spin_unlock(&dentry->d_lock); | |
2655 | ||
2656 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2657 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2658 | } | |
2659 | ||
2660 | /** | |
2661 | * d_materialise_unique - introduce an inode into the tree | |
2662 | * @dentry: candidate dentry | |
2663 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2664 | * | |
2665 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2666 | * root directory alias in its place if there is one. Caller must hold the |
2667 | * i_mutex of the parent directory. | |
770bfad8 DH |
2668 | */ |
2669 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2670 | { | |
9eaef27b | 2671 | struct dentry *actual; |
770bfad8 DH |
2672 | |
2673 | BUG_ON(!d_unhashed(dentry)); | |
2674 | ||
770bfad8 DH |
2675 | if (!inode) { |
2676 | actual = dentry; | |
360da900 | 2677 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2678 | d_rehash(actual); |
2679 | goto out_nolock; | |
770bfad8 DH |
2680 | } |
2681 | ||
873feea0 | 2682 | spin_lock(&inode->i_lock); |
357f8e65 | 2683 | |
9eaef27b TM |
2684 | if (S_ISDIR(inode->i_mode)) { |
2685 | struct dentry *alias; | |
2686 | ||
2687 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2688 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2689 | if (alias) { |
2690 | actual = alias; | |
18367501 AV |
2691 | write_seqlock(&rename_lock); |
2692 | ||
2693 | if (d_ancestor(alias, dentry)) { | |
2694 | /* Check for loops */ | |
2695 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2696 | spin_unlock(&inode->i_lock); |
18367501 AV |
2697 | } else if (IS_ROOT(alias)) { |
2698 | /* Is this an anonymous mountpoint that we | |
2699 | * could splice into our tree? */ | |
9eaef27b | 2700 | __d_materialise_dentry(dentry, alias); |
18367501 | 2701 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2702 | __d_drop(alias); |
2703 | goto found; | |
18367501 AV |
2704 | } else { |
2705 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2706 | * aliasing. This drops inode->i_lock */ |
18367501 | 2707 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2708 | } |
18367501 | 2709 | write_sequnlock(&rename_lock); |
dd179946 DH |
2710 | if (IS_ERR(actual)) { |
2711 | if (PTR_ERR(actual) == -ELOOP) | |
2712 | pr_warn_ratelimited( | |
2713 | "VFS: Lookup of '%s' in %s %s" | |
2714 | " would have caused loop\n", | |
2715 | dentry->d_name.name, | |
2716 | inode->i_sb->s_type->name, | |
2717 | inode->i_sb->s_id); | |
9eaef27b | 2718 | dput(alias); |
dd179946 | 2719 | } |
9eaef27b TM |
2720 | goto out_nolock; |
2721 | } | |
770bfad8 DH |
2722 | } |
2723 | ||
2724 | /* Add a unique reference */ | |
2725 | actual = __d_instantiate_unique(dentry, inode); | |
2726 | if (!actual) | |
2727 | actual = dentry; | |
357f8e65 NP |
2728 | else |
2729 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2730 | |
770bfad8 DH |
2731 | spin_lock(&actual->d_lock); |
2732 | found: | |
2733 | _d_rehash(actual); | |
2734 | spin_unlock(&actual->d_lock); | |
873feea0 | 2735 | spin_unlock(&inode->i_lock); |
9eaef27b | 2736 | out_nolock: |
770bfad8 DH |
2737 | if (actual == dentry) { |
2738 | security_d_instantiate(dentry, inode); | |
2739 | return NULL; | |
2740 | } | |
2741 | ||
2742 | iput(inode); | |
2743 | return actual; | |
770bfad8 | 2744 | } |
ec4f8605 | 2745 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2746 | |
cdd16d02 | 2747 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2748 | { |
2749 | *buflen -= namelen; | |
2750 | if (*buflen < 0) | |
2751 | return -ENAMETOOLONG; | |
2752 | *buffer -= namelen; | |
2753 | memcpy(*buffer, str, namelen); | |
2754 | return 0; | |
2755 | } | |
2756 | ||
232d2d60 WL |
2757 | /** |
2758 | * prepend_name - prepend a pathname in front of current buffer pointer | |
2759 | * buffer: buffer pointer | |
2760 | * buflen: allocated length of the buffer | |
2761 | * name: name string and length qstr structure | |
2762 | * | |
2763 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2764 | * make sure that either the old or the new name pointer and length are | |
2765 | * fetched. However, there may be mismatch between length and pointer. | |
2766 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2767 | * the length is reached or a null byte is found. It also prepends "/" at | |
2768 | * the beginning of the name. The sequence number check at the caller will | |
2769 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2770 | * due to mismatched pointer and length will be discarded. | |
2771 | */ | |
cdd16d02 MS |
2772 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2773 | { | |
232d2d60 WL |
2774 | const char *dname = ACCESS_ONCE(name->name); |
2775 | u32 dlen = ACCESS_ONCE(name->len); | |
2776 | char *p; | |
2777 | ||
2778 | if (*buflen < dlen + 1) | |
2779 | return -ENAMETOOLONG; | |
2780 | *buflen -= dlen + 1; | |
2781 | p = *buffer -= dlen + 1; | |
2782 | *p++ = '/'; | |
2783 | while (dlen--) { | |
2784 | char c = *dname++; | |
2785 | if (!c) | |
2786 | break; | |
2787 | *p++ = c; | |
2788 | } | |
2789 | return 0; | |
cdd16d02 MS |
2790 | } |
2791 | ||
1da177e4 | 2792 | /** |
208898c1 | 2793 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2794 | * @path: the dentry/vfsmount to report |
02125a82 | 2795 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2796 | * @buffer: pointer to the end of the buffer |
2797 | * @buflen: pointer to buffer length | |
552ce544 | 2798 | * |
232d2d60 WL |
2799 | * The function tries to write out the pathname without taking any lock other |
2800 | * than the RCU read lock to make sure that dentries won't go away. It only | |
2801 | * checks the sequence number of the global rename_lock as any change in the | |
2802 | * dentry's d_seq will be preceded by changes in the rename_lock sequence | |
2803 | * number. If the sequence number had been change, it will restart the whole | |
2804 | * pathname back-tracing sequence again. It performs a total of 3 trials of | |
2805 | * lockless back-tracing sequences before falling back to take the | |
2806 | * rename_lock. | |
1da177e4 | 2807 | */ |
02125a82 AV |
2808 | static int prepend_path(const struct path *path, |
2809 | const struct path *root, | |
f2eb6575 | 2810 | char **buffer, int *buflen) |
1da177e4 | 2811 | { |
9d1bc601 MS |
2812 | struct dentry *dentry = path->dentry; |
2813 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2814 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 | 2815 | int error = 0; |
232d2d60 WL |
2816 | unsigned seq = 0; |
2817 | char *bptr; | |
2818 | int blen; | |
6092d048 | 2819 | |
48f5ec21 | 2820 | rcu_read_lock(); |
232d2d60 WL |
2821 | restart: |
2822 | bptr = *buffer; | |
2823 | blen = *buflen; | |
2824 | read_seqbegin_or_lock(&rename_lock, &seq); | |
f2eb6575 | 2825 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2826 | struct dentry * parent; |
2827 | ||
1da177e4 | 2828 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2829 | /* Global root? */ |
232d2d60 WL |
2830 | if (mnt_has_parent(mnt)) { |
2831 | dentry = mnt->mnt_mountpoint; | |
2832 | mnt = mnt->mnt_parent; | |
2833 | vfsmnt = &mnt->mnt; | |
2834 | continue; | |
2835 | } | |
2836 | /* | |
2837 | * Filesystems needing to implement special "root names" | |
2838 | * should do so with ->d_dname() | |
2839 | */ | |
2840 | if (IS_ROOT(dentry) && | |
2841 | (dentry->d_name.len != 1 || | |
2842 | dentry->d_name.name[0] != '/')) { | |
2843 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2844 | (int) dentry->d_name.len, | |
2845 | dentry->d_name.name); | |
2846 | } | |
2847 | if (!error) | |
2848 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2849 | break; | |
1da177e4 LT |
2850 | } |
2851 | parent = dentry->d_parent; | |
2852 | prefetch(parent); | |
232d2d60 | 2853 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2854 | if (error) |
2855 | break; | |
2856 | ||
1da177e4 LT |
2857 | dentry = parent; |
2858 | } | |
48f5ec21 AV |
2859 | if (!(seq & 1)) |
2860 | rcu_read_unlock(); | |
2861 | if (need_seqretry(&rename_lock, seq)) { | |
2862 | seq = 1; | |
232d2d60 | 2863 | goto restart; |
48f5ec21 AV |
2864 | } |
2865 | done_seqretry(&rename_lock, seq); | |
1da177e4 | 2866 | |
232d2d60 WL |
2867 | if (error >= 0 && bptr == *buffer) { |
2868 | if (--blen < 0) | |
2869 | error = -ENAMETOOLONG; | |
2870 | else | |
2871 | *--bptr = '/'; | |
2872 | } | |
2873 | *buffer = bptr; | |
2874 | *buflen = blen; | |
7ea600b5 | 2875 | return error; |
f2eb6575 | 2876 | } |
be285c71 | 2877 | |
f2eb6575 MS |
2878 | /** |
2879 | * __d_path - return the path of a dentry | |
2880 | * @path: the dentry/vfsmount to report | |
02125a82 | 2881 | * @root: root vfsmnt/dentry |
cd956a1c | 2882 | * @buf: buffer to return value in |
f2eb6575 MS |
2883 | * @buflen: buffer length |
2884 | * | |
ffd1f4ed | 2885 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2886 | * |
2887 | * Returns a pointer into the buffer or an error code if the | |
2888 | * path was too long. | |
2889 | * | |
be148247 | 2890 | * "buflen" should be positive. |
f2eb6575 | 2891 | * |
02125a82 | 2892 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2893 | */ |
02125a82 AV |
2894 | char *__d_path(const struct path *path, |
2895 | const struct path *root, | |
f2eb6575 MS |
2896 | char *buf, int buflen) |
2897 | { | |
2898 | char *res = buf + buflen; | |
2899 | int error; | |
2900 | ||
2901 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2902 | br_read_lock(&vfsmount_lock); |
f2eb6575 | 2903 | error = prepend_path(path, root, &res, &buflen); |
7ea600b5 | 2904 | br_read_unlock(&vfsmount_lock); |
be148247 | 2905 | |
02125a82 AV |
2906 | if (error < 0) |
2907 | return ERR_PTR(error); | |
2908 | if (error > 0) | |
2909 | return NULL; | |
2910 | return res; | |
2911 | } | |
2912 | ||
2913 | char *d_absolute_path(const struct path *path, | |
2914 | char *buf, int buflen) | |
2915 | { | |
2916 | struct path root = {}; | |
2917 | char *res = buf + buflen; | |
2918 | int error; | |
2919 | ||
2920 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2921 | br_read_lock(&vfsmount_lock); |
02125a82 | 2922 | error = prepend_path(path, &root, &res, &buflen); |
7ea600b5 | 2923 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2924 | |
2925 | if (error > 1) | |
2926 | error = -EINVAL; | |
2927 | if (error < 0) | |
f2eb6575 | 2928 | return ERR_PTR(error); |
f2eb6575 | 2929 | return res; |
1da177e4 LT |
2930 | } |
2931 | ||
ffd1f4ed MS |
2932 | /* |
2933 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2934 | */ | |
02125a82 AV |
2935 | static int path_with_deleted(const struct path *path, |
2936 | const struct path *root, | |
2937 | char **buf, int *buflen) | |
ffd1f4ed MS |
2938 | { |
2939 | prepend(buf, buflen, "\0", 1); | |
2940 | if (d_unlinked(path->dentry)) { | |
2941 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2942 | if (error) | |
2943 | return error; | |
2944 | } | |
2945 | ||
2946 | return prepend_path(path, root, buf, buflen); | |
2947 | } | |
2948 | ||
8df9d1a4 MS |
2949 | static int prepend_unreachable(char **buffer, int *buflen) |
2950 | { | |
2951 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2952 | } | |
2953 | ||
a03a8a70 JB |
2954 | /** |
2955 | * d_path - return the path of a dentry | |
cf28b486 | 2956 | * @path: path to report |
a03a8a70 JB |
2957 | * @buf: buffer to return value in |
2958 | * @buflen: buffer length | |
2959 | * | |
2960 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2961 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2962 | * | |
52afeefb AV |
2963 | * Returns a pointer into the buffer or an error code if the path was |
2964 | * too long. Note: Callers should use the returned pointer, not the passed | |
2965 | * in buffer, to use the name! The implementation often starts at an offset | |
2966 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2967 | * |
31f3e0b3 | 2968 | * "buflen" should be positive. |
a03a8a70 | 2969 | */ |
20d4fdc1 | 2970 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2971 | { |
ffd1f4ed | 2972 | char *res = buf + buflen; |
6ac08c39 | 2973 | struct path root; |
ffd1f4ed | 2974 | int error; |
1da177e4 | 2975 | |
c23fbb6b ED |
2976 | /* |
2977 | * We have various synthetic filesystems that never get mounted. On | |
2978 | * these filesystems dentries are never used for lookup purposes, and | |
2979 | * thus don't need to be hashed. They also don't need a name until a | |
2980 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2981 | * below allows us to generate a name for these objects on demand: | |
2982 | */ | |
cf28b486 JB |
2983 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2984 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2985 | |
f7ad3c6b | 2986 | get_fs_root(current->fs, &root); |
7ea600b5 | 2987 | br_read_lock(&vfsmount_lock); |
02125a82 | 2988 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 | 2989 | br_read_unlock(&vfsmount_lock); |
02125a82 | 2990 | if (error < 0) |
ffd1f4ed | 2991 | res = ERR_PTR(error); |
6ac08c39 | 2992 | path_put(&root); |
1da177e4 LT |
2993 | return res; |
2994 | } | |
ec4f8605 | 2995 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2996 | |
c23fbb6b ED |
2997 | /* |
2998 | * Helper function for dentry_operations.d_dname() members | |
2999 | */ | |
3000 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3001 | const char *fmt, ...) | |
3002 | { | |
3003 | va_list args; | |
3004 | char temp[64]; | |
3005 | int sz; | |
3006 | ||
3007 | va_start(args, fmt); | |
3008 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3009 | va_end(args); | |
3010 | ||
3011 | if (sz > sizeof(temp) || sz > buflen) | |
3012 | return ERR_PTR(-ENAMETOOLONG); | |
3013 | ||
3014 | buffer += buflen - sz; | |
3015 | return memcpy(buffer, temp, sz); | |
3016 | } | |
3017 | ||
118b2302 AV |
3018 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3019 | { | |
3020 | char *end = buffer + buflen; | |
3021 | /* these dentries are never renamed, so d_lock is not needed */ | |
3022 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3023 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3024 | prepend(&end, &buflen, "/", 1)) |
3025 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3026 | return end; |
118b2302 AV |
3027 | } |
3028 | ||
6092d048 RP |
3029 | /* |
3030 | * Write full pathname from the root of the filesystem into the buffer. | |
3031 | */ | |
ec2447c2 | 3032 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 | 3033 | { |
232d2d60 WL |
3034 | char *end, *retval; |
3035 | int len, seq = 0; | |
3036 | int error = 0; | |
6092d048 | 3037 | |
48f5ec21 | 3038 | rcu_read_lock(); |
232d2d60 WL |
3039 | restart: |
3040 | end = buf + buflen; | |
3041 | len = buflen; | |
3042 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3043 | if (buflen < 1) |
3044 | goto Elong; | |
3045 | /* Get '/' right */ | |
3046 | retval = end-1; | |
3047 | *retval = '/'; | |
232d2d60 | 3048 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3049 | while (!IS_ROOT(dentry)) { |
3050 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 3051 | int error; |
6092d048 | 3052 | |
6092d048 | 3053 | prefetch(parent); |
232d2d60 WL |
3054 | error = prepend_name(&end, &len, &dentry->d_name); |
3055 | if (error) | |
3056 | break; | |
6092d048 RP |
3057 | |
3058 | retval = end; | |
3059 | dentry = parent; | |
3060 | } | |
48f5ec21 AV |
3061 | if (!(seq & 1)) |
3062 | rcu_read_unlock(); | |
3063 | if (need_seqretry(&rename_lock, seq)) { | |
3064 | seq = 1; | |
232d2d60 | 3065 | goto restart; |
48f5ec21 AV |
3066 | } |
3067 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3068 | if (error) |
3069 | goto Elong; | |
c103135c AV |
3070 | return retval; |
3071 | Elong: | |
3072 | return ERR_PTR(-ENAMETOOLONG); | |
3073 | } | |
ec2447c2 NP |
3074 | |
3075 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3076 | { | |
232d2d60 | 3077 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3078 | } |
3079 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3080 | |
3081 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3082 | { | |
3083 | char *p = NULL; | |
3084 | char *retval; | |
3085 | ||
c103135c AV |
3086 | if (d_unlinked(dentry)) { |
3087 | p = buf + buflen; | |
3088 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3089 | goto Elong; | |
3090 | buflen++; | |
3091 | } | |
3092 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3093 | if (!IS_ERR(retval) && p) |
3094 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3095 | return retval; |
3096 | Elong: | |
6092d048 RP |
3097 | return ERR_PTR(-ENAMETOOLONG); |
3098 | } | |
3099 | ||
1da177e4 LT |
3100 | /* |
3101 | * NOTE! The user-level library version returns a | |
3102 | * character pointer. The kernel system call just | |
3103 | * returns the length of the buffer filled (which | |
3104 | * includes the ending '\0' character), or a negative | |
3105 | * error value. So libc would do something like | |
3106 | * | |
3107 | * char *getcwd(char * buf, size_t size) | |
3108 | * { | |
3109 | * int retval; | |
3110 | * | |
3111 | * retval = sys_getcwd(buf, size); | |
3112 | * if (retval >= 0) | |
3113 | * return buf; | |
3114 | * errno = -retval; | |
3115 | * return NULL; | |
3116 | * } | |
3117 | */ | |
3cdad428 | 3118 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3119 | { |
552ce544 | 3120 | int error; |
6ac08c39 | 3121 | struct path pwd, root; |
552ce544 | 3122 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
3123 | |
3124 | if (!page) | |
3125 | return -ENOMEM; | |
3126 | ||
f7ad3c6b | 3127 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 3128 | |
552ce544 | 3129 | error = -ENOENT; |
7ea600b5 | 3130 | br_read_lock(&vfsmount_lock); |
f3da392e | 3131 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3132 | unsigned long len; |
8df9d1a4 MS |
3133 | char *cwd = page + PAGE_SIZE; |
3134 | int buflen = PAGE_SIZE; | |
1da177e4 | 3135 | |
8df9d1a4 | 3136 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3137 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
7ea600b5 | 3138 | br_read_unlock(&vfsmount_lock); |
552ce544 | 3139 | |
02125a82 | 3140 | if (error < 0) |
552ce544 LT |
3141 | goto out; |
3142 | ||
8df9d1a4 | 3143 | /* Unreachable from current root */ |
02125a82 | 3144 | if (error > 0) { |
8df9d1a4 MS |
3145 | error = prepend_unreachable(&cwd, &buflen); |
3146 | if (error) | |
3147 | goto out; | |
3148 | } | |
3149 | ||
552ce544 LT |
3150 | error = -ERANGE; |
3151 | len = PAGE_SIZE + page - cwd; | |
3152 | if (len <= size) { | |
3153 | error = len; | |
3154 | if (copy_to_user(buf, cwd, len)) | |
3155 | error = -EFAULT; | |
3156 | } | |
949854d0 | 3157 | } else { |
7ea600b5 | 3158 | br_read_unlock(&vfsmount_lock); |
949854d0 | 3159 | } |
1da177e4 LT |
3160 | |
3161 | out: | |
6ac08c39 JB |
3162 | path_put(&pwd); |
3163 | path_put(&root); | |
1da177e4 LT |
3164 | free_page((unsigned long) page); |
3165 | return error; | |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3170 | * | |
3171 | * Trivially implemented using the dcache structure | |
3172 | */ | |
3173 | ||
3174 | /** | |
3175 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3176 | * @new_dentry: new dentry | |
3177 | * @old_dentry: old dentry | |
3178 | * | |
3179 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3180 | * Returns 0 otherwise. | |
3181 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3182 | */ | |
3183 | ||
e2761a11 | 3184 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3185 | { |
3186 | int result; | |
949854d0 | 3187 | unsigned seq; |
1da177e4 | 3188 | |
e2761a11 OH |
3189 | if (new_dentry == old_dentry) |
3190 | return 1; | |
3191 | ||
e2761a11 | 3192 | do { |
1da177e4 | 3193 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3194 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3195 | /* |
3196 | * Need rcu_readlock to protect against the d_parent trashing | |
3197 | * due to d_move | |
3198 | */ | |
3199 | rcu_read_lock(); | |
e2761a11 | 3200 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3201 | result = 1; |
e2761a11 OH |
3202 | else |
3203 | result = 0; | |
949854d0 | 3204 | rcu_read_unlock(); |
1da177e4 | 3205 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3206 | |
3207 | return result; | |
3208 | } | |
3209 | ||
db14fc3a | 3210 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3211 | { |
db14fc3a MS |
3212 | struct dentry *root = data; |
3213 | if (dentry != root) { | |
3214 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3215 | return D_WALK_SKIP; | |
1da177e4 | 3216 | |
01ddc4ed MS |
3217 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3218 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3219 | dentry->d_lockref.count--; | |
3220 | } | |
1da177e4 | 3221 | } |
db14fc3a MS |
3222 | return D_WALK_CONTINUE; |
3223 | } | |
58db63d0 | 3224 | |
db14fc3a MS |
3225 | void d_genocide(struct dentry *parent) |
3226 | { | |
3227 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3228 | } |
3229 | ||
60545d0d | 3230 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3231 | { |
60545d0d AV |
3232 | inode_dec_link_count(inode); |
3233 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3234 | !hlist_unhashed(&dentry->d_alias) || | |
3235 | !d_unlinked(dentry)); | |
3236 | spin_lock(&dentry->d_parent->d_lock); | |
3237 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3238 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3239 | (unsigned long long)inode->i_ino); | |
3240 | spin_unlock(&dentry->d_lock); | |
3241 | spin_unlock(&dentry->d_parent->d_lock); | |
3242 | d_instantiate(dentry, inode); | |
1da177e4 | 3243 | } |
60545d0d | 3244 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3245 | |
3246 | static __initdata unsigned long dhash_entries; | |
3247 | static int __init set_dhash_entries(char *str) | |
3248 | { | |
3249 | if (!str) | |
3250 | return 0; | |
3251 | dhash_entries = simple_strtoul(str, &str, 0); | |
3252 | return 1; | |
3253 | } | |
3254 | __setup("dhash_entries=", set_dhash_entries); | |
3255 | ||
3256 | static void __init dcache_init_early(void) | |
3257 | { | |
074b8517 | 3258 | unsigned int loop; |
1da177e4 LT |
3259 | |
3260 | /* If hashes are distributed across NUMA nodes, defer | |
3261 | * hash allocation until vmalloc space is available. | |
3262 | */ | |
3263 | if (hashdist) | |
3264 | return; | |
3265 | ||
3266 | dentry_hashtable = | |
3267 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3268 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3269 | dhash_entries, |
3270 | 13, | |
3271 | HASH_EARLY, | |
3272 | &d_hash_shift, | |
3273 | &d_hash_mask, | |
31fe62b9 | 3274 | 0, |
1da177e4 LT |
3275 | 0); |
3276 | ||
074b8517 | 3277 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3278 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3279 | } |
3280 | ||
74bf17cf | 3281 | static void __init dcache_init(void) |
1da177e4 | 3282 | { |
074b8517 | 3283 | unsigned int loop; |
1da177e4 LT |
3284 | |
3285 | /* | |
3286 | * A constructor could be added for stable state like the lists, | |
3287 | * but it is probably not worth it because of the cache nature | |
3288 | * of the dcache. | |
3289 | */ | |
0a31bd5f CL |
3290 | dentry_cache = KMEM_CACHE(dentry, |
3291 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3292 | |
3293 | /* Hash may have been set up in dcache_init_early */ | |
3294 | if (!hashdist) | |
3295 | return; | |
3296 | ||
3297 | dentry_hashtable = | |
3298 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3299 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3300 | dhash_entries, |
3301 | 13, | |
3302 | 0, | |
3303 | &d_hash_shift, | |
3304 | &d_hash_mask, | |
31fe62b9 | 3305 | 0, |
1da177e4 LT |
3306 | 0); |
3307 | ||
074b8517 | 3308 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3309 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3310 | } |
3311 | ||
3312 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3313 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3314 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3315 | |
1da177e4 LT |
3316 | EXPORT_SYMBOL(d_genocide); |
3317 | ||
1da177e4 LT |
3318 | void __init vfs_caches_init_early(void) |
3319 | { | |
3320 | dcache_init_early(); | |
3321 | inode_init_early(); | |
3322 | } | |
3323 | ||
3324 | void __init vfs_caches_init(unsigned long mempages) | |
3325 | { | |
3326 | unsigned long reserve; | |
3327 | ||
3328 | /* Base hash sizes on available memory, with a reserve equal to | |
3329 | 150% of current kernel size */ | |
3330 | ||
3331 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3332 | mempages -= reserve; | |
3333 | ||
3334 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3335 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3336 | |
74bf17cf DC |
3337 | dcache_init(); |
3338 | inode_init(); | |
1da177e4 | 3339 | files_init(mempages); |
74bf17cf | 3340 | mnt_init(); |
1da177e4 LT |
3341 | bdev_cache_init(); |
3342 | chrdev_init(); | |
3343 | } |